diff --git a/docs/llmservice/api/API.md b/docs/llmservice/api/API.md
index ba141cb..90ec7a4 100644
--- a/docs/llmservice/api/API.md
+++ b/docs/llmservice/api/API.md
@@ -138,7 +138,9 @@ The request structure of the Responses API differs from Chat Completions:
 
 ### Model and Endpoint Compatibility
 
-The Responses endpoint accepts models enabled for this protocol. If the model and endpoint are incompatible, the server returns HTTP `400` with error details.
+The Responses endpoint supports GPT and DeepSeek model families available through B.AI. Use `GET /v1/models` to retrieve the model IDs available to the current API credential. Supported parameters and tools vary by model; if the model and endpoint are incompatible, the server returns HTTP `400` with error details.
+
+DeepSeek models do not support web search. When calling a DeepSeek model through the Responses API, do not include a web search tool in the request. When using a DeepSeek model in Codex, set the top-level Codex option `web_search = "disabled"`.
 
 ### Request Body
 
@@ -669,7 +671,7 @@ When `stream: true`, common events include:
 
 ## Codex CLI Integration
 
-The B.AI Responses API can be used as a custom model provider for Codex. The following configuration applies to Codex versions that support custom model providers.
+The B.AI Responses API can be used as a custom model provider for Codex with supported GPT and DeepSeek model families. The following configuration applies to Codex versions that support custom model providers.
 
 ### 1. Set the API Key
 
@@ -699,6 +701,20 @@ wire_api = "responses"
 requires_openai_auth = false
 ```
 
+:::caution Disable Codex web search before using DeepSeek
+Codex may expose its built-in web search tool by default, but DeepSeek models do not support that tool. Without this setting, the request may return an unsupported Web Search tool error.
+
+Open `~/.codex/config.toml` and add `web_search = "disabled"` as a top-level option:
+
+```toml
+model = "your-model-id"
+model_provider = "bai"
+web_search = "disabled"
+```
+
+Keep `web_search` above and outside the `[model_providers.bai]` block. Save the file, fully quit Codex, and restart it. This setting is required when using DeepSeek; remove or change it when switching to a model configuration that uses web search.
+:::
+
 If the file already contains configuration, append the complete `[model_providers.bai]` block and change the top-level `model` and `model_provider` values as shown above. See the Codex documentation at the end of this page for a full explanation of the configuration fields.
 
 After saving, start Codex from a terminal where `BAI_API_KEY` is set:
@@ -722,6 +738,7 @@ model = "your-model-id"
 | `401` response | Check whether the API Key is valid and whether a Key from another environment was used accidentally. |
 | `403` response | Check the account status and model permissions. |
 | Model not supported | Confirm that the model ID is spelled correctly and is enabled for the configured endpoint. |
+| DeepSeek request reports an unsupported web search tool | Confirm that the top-level configuration contains `web_search = "disabled"`, then restart Codex. |
 
 ---
 
@@ -788,6 +805,7 @@ Error responses include an HTTP status code and an `error` object. Applications
 | Scenario | Status code | Handling |
 |---|---:|---|
 | Model and endpoint are incompatible | `400` | Select a model enabled for the endpoint or use another endpoint. |
+| A DeepSeek request includes web search | `400` | Remove the web search tool. In Codex, set the top-level option `web_search = "disabled"`. |
 | `max_tokens` or `max_completion_tokens` is used | `400` | Use `max_output_tokens` instead. |
 | `max_output_tokens` exceeds the model limit | `400` | Adjust the value to the range stated in the error. |
 | Request uses an unavailable tool | `400` | Remove the tool or choose a compatible model configuration. |
diff --git a/docs/llmservice/models/claude-fable-5-1.md b/docs/llmservice/models/claude-fable-5-1.md
new file mode 100644
index 0000000..d5f7eb5
--- /dev/null
+++ b/docs/llmservice/models/claude-fable-5-1.md
@@ -0,0 +1,56 @@
+# Claude Fable 5.1
+
+## Overview
+
+Claude Fable 5.1, released by Anthropic on September 1, 2026, is a generally available Mythos-class model in the Claude 5 family for demanding reasoning and long-horizon agentic work. It shares its underlying model with the limited-access Claude Mythos 5.1. On B.AI, use the model ID `claude-fable-5-1`.
+
+## Key Features
+
+* **Long-Horizon Agentic Performance:** Anthropic reports 52.6% on Terminal-Bench-Science 0.1, 55.8% on Terminal-Bench 4.0, 31.4% on AutomationBench, and 73.4% on CursorBench 3.2.0 with production safeguards enabled.
+* **Always-On Adaptive Thinking:** Adaptive thinking cannot be disabled. The `effort` control supports `low`, `medium`, `high`, `xhigh`, and `max`, with `high` as the Claude API default.
+* **1M Context and 128K Output:** Provides a 1,000,000-token context window and up to 128,000 output tokens at standard per-token pricing across the full window.
+* **Agent-Oriented API Controls:** Supports automatic tool use, strict tool schemas, structured output, preserved thinking, per-message effort, turn-scoped system messages, readable progress updates, and content provenance. Several of these controls are beta features.
+* **Lower Cache-Read Cost:** Cache hits and refreshes cost 0.25 Credits/Token, one quarter of the Claude Fable 5 cache-read rate, while base input and output prices remain unchanged.
+
+## Best Use Cases
+
+* **Long-Running Software Engineering:** Repository-scale implementation, debugging, code review, performance work, and autonomous sessions that span many tools and checkpoints.
+* **Multi-Step Research and Analysis:** Research workflows that require search, evidence synthesis, iterative hypothesis testing, and recovery from failed steps.
+* **Enterprise Knowledge Work:** Complex document, spreadsheet, presentation, financial-analysis, and cross-file workflows that benefit from a large context window and vision.
+* **Browser and Computer-Use Agents:** Tasks that operate across applications, maintain state over extended sessions, and provide progress updates between tool calls.
+
+## Capabilities and Limitations
+
+| Capability | Description |
+| :--- | :--- |
+| **Reasoning** | Always-on adaptive thinking with five effort levels. Anthropic reports 60.9% on Humanity's Last Exam without tools and 65.0% with tools. |
+| **Creative Writing** | Supports drafting, editing, research synthesis, and professional artifact creation. |
+| **Multimodal** | Accepts text and image input and produces text output. It can analyze charts, diagrams, tables, files, and PDFs. |
+| **Context Window** | 1M tokens. |
+| **Max Output** | 128,000 tokens. |
+| **Tool Use** | Supports automatic function calling, strict tool use, structured output, browser and computer-use workflows, and preserved thinking. Forced `any` or named-tool selection is not supported. |
+| **Multilingual** | Anthropic lists multilingual support across current Claude models. |
+| **Knowledge Cutoff** | Reliable knowledge cutoff and training-data cutoff: June 2026. |
+
+### Known Limitations
+
+* Forced tool choice is not supported. Setting `tool_choice` to `any` or to a named tool returns a 400 error; integrations should use `auto` or `none` and express required tool behavior in the prompt.
+* Thinking blocks are forward-compatible only: Claude Fable 5.1 can read thinking blocks from earlier models, but earlier models cannot read its thinking blocks.
+* Editing the system prompt, tool definitions, or earlier messages after a Fable 5.1 thinking block can invalidate that block. Long-running integrations should treat conversation history as append-only or use the documented block-dropping controls.
+* The model is comparatively slow and costs twice as much per base input and output token as Claude Opus 5.
+* Fable safeguards may refuse or route some cybersecurity and biology requests to another Claude model. Fable 5.1 also requires 30-day data retention by default; zero-data-retention use requires express authorization from Anthropic.
+* Priority Tier is not supported for Claude Fable 5.1.
+
+## Pricing
+
+| Model | Input (Credits/Token) | 5m Cache Write (Credits/Token) | 1h Cache Write (Credits/Token) | Cache Read (Credits/Token) | Output (Credits/Token) | Web Search (Credits/Use) |
+| :--- | --------------------: | -----------------------------: | -----------------------------: | -------------------------: | ---------------------: | -----------------------: |
+| **Claude Fable 5.1** | `10.00` | `12.50` | `20.00` | `0.25` | `50.00` | `-` |
+
+:::info Caching note
+For Claude Fable 5.1, a 5-minute cache write is billed at 1.25x the input rate and a 1-hour cache write is billed at 2x. Cache reads are billed at 0.025x the input rate.
+:::
+
+:::info Pricing note
+Prices shown in the documentation are B.AI standard reference prices for base billing purposes. B.AI may provide lower actual usage costs through top-up bonuses and account benefits. Specific prices, bonus Credits, and account benefits are subject to the platform display and final billing records.
+:::
diff --git a/docs/llmservice/models/deepseek-v4-flash-vision-exp.md b/docs/llmservice/models/deepseek-v4-flash-vision-exp.md
new file mode 100644
index 0000000..c29e79d
--- /dev/null
+++ b/docs/llmservice/models/deepseek-v4-flash-vision-exp.md
@@ -0,0 +1,67 @@
+import ActivityCard from '@site/src/components/ActivityCard';
+
+# DeepSeek-V4-Flash-Vision-Exp
+
+<ActivityCard
+  variant="discount"
+  title="DeepSeek-V4-Flash-Vision-Exp"
+  status="Limited-Time API Discount"
+  detail="50% of Standard Price"
+>
+The DeepSeek-V4-Flash-Vision-Exp 50% offer takes effect at 17:00 on September 3, 2026 (UTC+8).
+
+From the effective time, eligible `deepseek-v4-flash-vision-exp` API usage is billed at 50% of the standard price for the applicable Idle or Busy period. The discounted price changes in step with DeepSeek's time-based pricing and remains at 50% in either period.
+
+The pricing table below continues to show standard reference prices. Actual settlement and final billing are subject to the platform display.
+</ActivityCard>
+
+## Overview
+
+DeepSeek-V4-Flash-Vision-Exp is DeepSeek's first experimental multimodal model in the DeepSeek-V4 family, released on August 21, 2026. It extends DeepSeek-V4-Flash with visual modules and continued training for image understanding while retaining comparable text-only agent performance in DeepSeek's evaluations.
+
+## Key Features
+
+- **Native Image Understanding**: Accepts interleaved text and image input in JPEG, PNG, GIF, and WebP formats for screenshot reading, chart analysis, visual question answering, and other image-grounded tasks.
+- **Multimodal Agent Evaluation**: DeepSeek reports 36.5 on ApexBench (Pass@1), 64.3 on Chartography, and 35.0 on ZeroBench (Pass@5), indicating stronger visually grounded agent performance than the text-only DeepSeek-V4-Flash-0731 baseline in its published evaluation.
+- **Text-Agent Continuity**: DeepSeek reports text-only capabilities comparable to DeepSeek-V4-Flash, including 83.9 on Terminal-Bench 2.1, 59.3 on DeepSWE, and 75.9 on Toolathlon-Verified.
+- **Long-Context Reasoning**: The hosted API provides a 1M-token context window, up to 384K output tokens, optional thinking, and `low`, `high`, or `max` reasoning effort.
+- **Agent-Oriented API Support**: Supports tool calls, JSON output, context caching, Chat Prefix Completion, and OpenAI-compatible Chat Completions and Responses interfaces, as well as an Anthropic-compatible Messages interface.
+- **Open-Weight Reference Release**: The official MIT-licensed repository lists a 305B-parameter checkpoint and includes reference code for the vision encoder and aligner, DFlash attention, mixture-of-experts layers, Hyper-Connections, and the DSpark forward path.
+
+## Best Use Cases
+
+- **Visual Software Agents**: Inspecting screenshots, rendered interfaces, dashboards, or canvas state while using tools to modify code or other artifacts.
+- **Document and Chart Analysis**: Reading text from screenshots, interpreting charts, and answering questions grounded in visual business or technical material.
+- **Multimodal Tool Workflows**: Processing images returned by tools through the Responses API and combining visual evidence with multi-step reasoning and function calls.
+- **Large Visual Review Batches**: Handling many images in one request when applications need consistent analysis across collections of screenshots, pages, or frames.
+- **Open-Weight Evaluation**: Testing DeepSeek's experimental V4 vision architecture in controlled self-hosted research environments where the large checkpoint and reference-runtime requirements are acceptable.
+
+## Capabilities and Limitations
+
+| Capability | Description |
+| :--- | :--- |
+| **Reasoning** | Supports thinking and non-thinking modes; thinking is enabled by default at `high` effort. The primary effort levels are `low`, `high`, and `max`; compatible `medium` and `xhigh` requests map to `high`. |
+| **Coding and Agents** | DeepSeek reports Terminal-Bench 2.1: 83.9, DeepSWE: 59.3, and Toolathlon-Verified: 75.9. These are provider-run evaluations using DeepSeek Harness minimal mode, `max` effort, `temperature = 1.0`, and `top_p = 0.95`. |
+| **Tool Use** | Supports function calling in thinking and non-thinking modes, JSON output, context caching, and image-bearing tool outputs through the Responses API. |
+| **Multimodal** | Accepts text and image input and produces text output. Each image is resized before inference and uses at most 384 input tokens; the API accepts up to 600 images per request within its size limits. |
+| **Context Window** | 1M tokens. |
+| **Max Output** | 384K tokens. |
+
+### Known Limitations
+
+- Image placement is interface-dependent: Chat Completions accepts images in `user` messages, while the Responses API also accepts them in `developer` messages and supported tool outputs. Images in `system` or `assistant` messages are rejected.
+- The documented multimodal route accepts images but does not document audio or video input, image generation, or other non-text output.
+- FIM Completion is not supported. Chat Prefix Completion remains a beta feature.
+- In thinking mode, `temperature`, `top_p`, `presence_penalty`, and `frequency_penalty` have no effect. Tool-calling conversations must pass the full prior `reasoning_content` back on subsequent requests.
+- DeepSeek does not publish a model-specific knowledge cutoff or a complete supported-language list.
+
+## Pricing
+
+| Billing Period | Input (Credits/Token) | Cache Write (Credits/Token) | Cache Read (Credits/Token) | Output (Credits/Token) |
+| :--- | ---: | ---: | ---: | ---: |
+| **Idle** | 0.22 | 0.22 | 0.0073 | 0.66 |
+| **Busy** | 0.44 | 0.44 | 0.0147 | 1.32 |
+
+API calls use UTC+8. Busy periods are 09:00-12:00 and 14:00-18:00, Monday through Friday; all other times, including weekends, are Idle periods.
+
+Images are converted to input tokens according to their dimensions and billed together with text input; each image uses at most 384 tokens after resizing.
diff --git a/docs/llmservice/models/deepseek-v4-flash.md b/docs/llmservice/models/deepseek-v4-flash.md
index 3283673..9799553 100644
--- a/docs/llmservice/models/deepseek-v4-flash.md
+++ b/docs/llmservice/models/deepseek-v4-flash.md
@@ -7,17 +7,16 @@ import ActivityCard from '@site/src/components/ActivityCard';
 DeepSeek-V4-Flash is DeepSeek's high-efficiency open-source language model, released alongside V4-Pro on April 24, 2026 under the MIT License. With 284 billion total parameters and only 13 billion active parameters, it delivers performance within striking distance of V4-Pro at roughly 3.1x lower cost, making it one of the most cost-effective models available.
 
 <ActivityCard
-  variant="free"
+  variant="discount"
   title="DeepSeek-V4-Flash"
-  status="Free Offer"
-  detail="0 Credits"
+  status="Limited-Time API Discount"
+  detail="50% of Standard Price"
 >
-Offer starts August 17, 2026, and applies to DeepSeek-V4-Flash on B.AI Chat and API.
+The DeepSeek-V4-Flash 50% offer takes effect at 17:00 on September 3, 2026 (UTC+8).
 
-* **Chat:** Usage is billed at `0 Credits` during the offer.
-* **API:** Usage is billed at `0 Credits` during the offer. No per-request, input, cache write, cache read, or output token fees apply.
+From the effective time, eligible DeepSeek-V4-Flash API usage is billed at 50% of the standard price for the applicable period. The discounted price changes in step with DeepSeek's Idle and Busy pricing periods and remains at 50% in either period.
 
-After the offer ends, the model will return to standard pricing. Offer end time, eligibility, actual charges, and final billing are subject to the platform display.
+The pricing table below continues to show standard reference prices. Actual settlement and final billing are subject to the platform display.
 </ActivityCard>
 
 ## Key Features
@@ -60,5 +59,5 @@ After the offer ends, the model will return to standard pricing. Offer end time,
 | **Busy** | `0.44` | `0.44` | `0.0147` | `1.32` | `-` | Cache Write: `1x` input; Cache Read: `0.0333x` input |
 
 :::info Pricing note
-The table shows the time-based standard reference price for DeepSeek-V4-Flash. Under DeepSeek's updated rule, from 00:00 UTC+8 on August 23, 2026, Saturdays and Sundays are billed at Idle rates all day. Its current limited-time offer applies `0 Credits` to all B.AI Chat and API usage. After the offer ends, the applicable weekday Idle or Busy period, final settlement price, and billing records are subject to the platform display. B.AI may provide lower actual usage costs through top-up bonuses and account benefits.
+The table shows the time-based standard reference price for DeepSeek-V4-Flash. API calls use UTC+8: Busy periods are 09:00-12:00 and 14:00-18:00, Monday through Friday; all other times, including weekends, are Idle periods. DeepSeek-V4-Flash usage in B.AI Chat is billed at Idle rates. From 17:00 on September 3, 2026 (UTC+8), eligible API usage is billed at 50% of the standard price for the applicable period. Final settlement prices and billing records are subject to the platform display. B.AI may provide lower actual usage costs through top-up bonuses and account benefits.
 :::
diff --git a/docs/llmservice/models/deepseek-v4-pro.md b/docs/llmservice/models/deepseek-v4-pro.md
index 1db8197..d0a78fb 100644
--- a/docs/llmservice/models/deepseek-v4-pro.md
+++ b/docs/llmservice/models/deepseek-v4-pro.md
@@ -44,5 +44,5 @@ DeepSeek-V4-Pro is DeepSeek's flagship open-source large language model, release
 | **Busy** | `1.32` | `1.32` | `0.044` | `3.96` | `-` | Cache Write: `1x` input; Cache Read: `0.0333x` input |
 
 :::info Pricing note
-DeepSeek V4 Pro uses time-based standard reference pricing. Under DeepSeek's updated rule, from 00:00 UTC+8 on August 23, 2026, Saturdays and Sundays are billed at Idle rates all day. The applicable weekday Idle or Busy period, final settlement price, and billing records are subject to the platform display. B.AI may provide lower actual usage costs through top-up bonuses and account benefits.
+DeepSeek V4 Pro uses time-based standard reference pricing. API calls use UTC+8: Busy periods are 09:00-12:00 and 14:00-18:00, Monday through Friday; all other times, including weekends, are Idle periods. DeepSeek V4 Pro usage in B.AI Chat is billed at Idle rates. Final settlement prices and billing records are subject to the platform display. B.AI may provide lower actual usage costs through top-up bonuses and account benefits.
 :::
diff --git a/docs/llmservice/models/gemini-3-8-flash.md b/docs/llmservice/models/gemini-3-8-flash.md
new file mode 100644
index 0000000..4d5ea91
--- /dev/null
+++ b/docs/llmservice/models/gemini-3-8-flash.md
@@ -0,0 +1,47 @@
+# Gemini 3.8 Flash
+
+## Overview
+
+Gemini 3.8 Flash is a generally available Gemini 3-series model released by Google on September 2, 2026. It is positioned for long-horizon software engineering, autonomous agents, and complex knowledge workflows, with a 1M-token input window and configurable reasoning effort.
+
+## Key Features
+
+- **Long-Horizon Agentic Execution**: Designed for multi-step planning, iterative tool use, and verification across extended software-engineering and enterprise workflows.
+- **Configurable Thinking**: Supports `low`, `medium`, and `high` thinking levels; `medium` is the default, while `minimal` is rejected.
+- **Native Multimodal Input**: Accepts text, images, video, audio, and PDFs within a 1,048,576-token input limit and produces up to 65,536 text tokens.
+- **Broad Tool Support**: Supports function calling, structured outputs, code execution, File Search, URL context, Search and Maps grounding, and Computer Use in preview.
+
+## Best Use Cases
+
+- **Agentic Software Engineering**: Repository analysis, multi-file implementation, debugging, and migration workflows that require repeated reasoning and tool calls.
+- **Complex Knowledge Work**: Long-form analysis and report generation across large document sets, structured data, and domain-specific evidence.
+- **Multimodal Analysis**: Understanding documents, screenshots, charts, video, and audio together with text instructions.
+- **Tool-Orchestrated Applications**: Agents that combine custom functions with code execution, search, URL retrieval, file search, or computer interaction.
+- **Asynchronous Bulk Processing**: Classification, extraction, evaluation, and other delay-tolerant workloads that can use Batch or Flex inference.
+
+## Capabilities and Limitations
+
+| Capability | Description |
+| :--- | :--- |
+| **Reasoning** | Supports `low`, `medium`, and `high` thinking levels, with `medium` as the default. Higher effort can improve difficult multi-step work but may increase token consumption and latency. |
+| **Coding** | Optimized for long-horizon software-engineering and agentic coding workflows. Google's release materials report gains over Gemini 3.7 Flash; provider-run evaluations are not production guarantees. |
+| **Multimodal** | Accepts text, image, video, audio, and PDF input and produces text output. |
+| **Context Window** | 1,048,576 input tokens. |
+| **Max Output** | 65,536 tokens. |
+| **Tool Use** | Function calling, structured outputs, code execution, File Search, URL context, Search and Maps grounding, and Computer Use (Preview). |
+| **Knowledge Cutoff** | March 2026 overall; Google notes that knowledge in some domains may be limited to January 2025. Use grounding for newer or time-sensitive information. |
+
+### Known Limitations
+
+- The model can hallucinate, and Google notes that occasional slowness or timeout issues may occur.
+- Difficult tasks and higher thinking levels may consume more reasoning and output tokens than earlier Flash workflows; applications should tune effort to their latency and cost targets.
+- `minimal` thinking is unsupported. Legacy sampling parameters including `temperature`, `top_p`, and `top_k` are ignored, and prefilled model turns are rejected.
+- Computer Use remains a preview capability, and the model does not support native image generation or native audio generation.
+
+## Pricing
+
+| Model | Input (Credits/Token) | Cache Write (Credits/Token) | Cache Read (Credits/Token) | Output (Credits/Token) | Web Search (Credits/Use) |
+| :--- | ---: | ---: | ---: | ---: | ---: |
+| Gemini 3.8 Flash | 0.75 | 0.75 | 0.075 | 3.75 | 14,000 |
+
+Explicit cache storage costs 0.50 Credits/Token per hour through December 31, 2026, and 1.00 Credits/Token per hour starting January 1, 2027.
diff --git a/docs/llmservice/models/hy4-preview.md b/docs/llmservice/models/hy4-preview.md
new file mode 100644
index 0000000..6aa6a61
--- /dev/null
+++ b/docs/llmservice/models/hy4-preview.md
@@ -0,0 +1,53 @@
+# Hy4 Preview
+
+## Overview
+
+Hy4 Preview is an open-weight Mixture-of-Experts language model released by the Tencent Hy Team on August 28, 2026. It is an early Hy4-series release with 770 billion backbone parameters, 49 billion activated parameters per token, and a 1M-token context window. On B.AI, use the model ID `hy4-preview`.
+
+## Key Features
+
+* **770B Sparse MoE Architecture:** Uses 78 backbone layers, 256 routed experts and one shared expert per MoE layer, and activates the top eight routed experts plus the shared expert for each token. A separate native MTP layer adds 10B total and 0.7B activated parameters for speculative decoding.
+* **Long-Context API Limits:** Supports a 1M-token context window, up to 960K input tokens, and up to 64K output tokens.
+* **Preserved Deep Reasoning:** Thinking is enabled by default. The API exposes `reasoning_content`, which should be returned with the complete assistant message during multi-turn tool workflows.
+* **Agent API Support:** Supports structured output, function calling, cached input, and OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages interfaces.
+* **Productivity-Oriented Validation:** In Tencent's blind internal evaluation, 163 experts rated outputs on 203 engineering tasks. Hy4 Preview averaged 2.99/4.00, compared with 2.92 for GLM-5.3 and 2.94 for Kimi K3.
+
+## Best Use Cases
+
+* **Software Engineering Agents:** Long-horizon planning, implementation, debugging, verification, frontend work, and multi-step tool use across large repositories.
+* **Office and Data Workflows:** Cross-document analysis, financial modeling, data analysis, and production of documents, spreadsheets, and presentations.
+* **Game Development:** Playable-prototype generation, game-engine workflows, and iterative refinement over multiple turns.
+* **Scientific Research:** Tool-assisted analysis and problem solving in areas such as AI research, molecular dynamics, condensed-matter physics, and mathematics.
+* **Private or Customized Deployment:** Open-weight deployments that require local serving, quantization, or task-specific fine-tuning and can accommodate the model's substantial compute requirements.
+
+## Capabilities and Limitations
+
+| Capability | Description |
+| :--- | :--- |
+| **Reasoning** | Supports deep reasoning with preserved thinking. The documented reasoning-effort levels are `low` and `high`, with `high` as the default for `hy4-preview` tool workflows. |
+| **Creative Writing** | Supports general text generation and Chinese content creation. |
+| **Multimodal** | Text input and text output. |
+| **Response Speed** | Tencent has not published a guaranteed generation rate. The official release reports a 31.8% throughput gain from model-assisted inference-system optimization relative to Tencent's internal baseline. |
+| **Context Window** | 1M tokens. |
+| **Maximum Input** | 960,000 tokens. |
+| **Max Output** | 64,000 tokens. |
+| **Tool Use** | Function calling, structured output, automatic tool selection, cached input, and preserved reasoning across tool calls. |
+| **Multilingual** | Tencent documents Chinese and English use. |
+
+### Known Limitations
+
+* Tencent identifies this release as an early version of Hy4 rather than a final Hy4 model.
+* The official model card reports that Hy4 Preview can spend longer than necessary reasoning through complex tasks and can over-verify its own work.
+* Multi-turn tool integrations should preserve and replay the full assistant message, including `reasoning_content`; omitting it can degrade the documented continuation workflow.
+* Hy4 Preview does not support native image, audio, or video understanding.
+* Tencent has not published a model-specific knowledge cutoff, guaranteed API throughput, or complete language-support list.
+
+## Pricing
+
+| Model | Input (Credits/Token) | Cache Write (Credits/Token) | Cache Read (Credits/Token) | Output (Credits/Token) | Web Search (Credits/Use) |
+| :--- | --------------------: | --------------------------: | -------------------------: | ---------------------: | -----------------------: |
+| **Hy4 Preview** | `0.834` | `0.834` | `0.042` | `2.501` | `-` |
+
+:::info Pricing note
+Prices shown in the documentation are B.AI standard reference prices for base billing purposes. B.AI may provide lower actual usage costs through top-up bonuses and account benefits. Specific prices, bonus Credits, and account benefits are subject to the platform display and final billing records.
+:::
diff --git a/docs/llmservice/models/muse-spark-1-3.md b/docs/llmservice/models/muse-spark-1-3.md
new file mode 100644
index 0000000..dea1626
--- /dev/null
+++ b/docs/llmservice/models/muse-spark-1-3.md
@@ -0,0 +1,42 @@
+# Muse Spark 1.3
+
+## Overview
+
+Muse Spark 1.3 is a proprietary agentic and coding model from Meta Superintelligence Labs, released on September 2, 2026. Meta makes it available through Muse Code and Meta Model API and positions the update for longer-horizon work, stronger instruction retention, improved multitasking, and more efficient coding-agent execution.
+
+## Key Features
+
+- **Long-Horizon Agent Work**: Designed to build context with tools, identify gaps in its plan, retain learned information, and complete open-ended objectives across extended workflows.
+- **Collaborative Control**: Trained to ask clarifying questions when instructions are ambiguous, request help when blocked, and confirm before consequential actions.
+- **Multi-Workflow Tracking**: Improved at mapping new prompts to the correct task in long, interrupted, or mixed-purpose threads without dropping earlier requirements.
+- **Coding Efficiency**: In comparisons run by Meta engineers, the model used about 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2 while producing cleaner, less verbose code.
+- **Improved Safety Calibration**: Meta reports stronger resistance to prompt injection and better judgment around irreversible actions in agentic workflows.
+
+## Best Use Cases
+
+- **Long-Running Coding Agents**: Feature implementation, debugging, codebase analysis, and other engineering tasks that need iterative tools and sustained instruction following.
+- **Complex Deliverable Workflows**: Producing reports, presentations, analyses, or other artifacts from multiple files and conflicting source materials.
+- **Research and Automation Agents**: Workflows that gather their own context, revise plans, and combine browsing or business-automation tools.
+- **Multi-Task Collaboration**: Persistent assistant threads where users interrupt, redirect, or resume several related workstreams.
+
+## Capabilities and Limitations
+
+| Capability | Description |
+| :--- | :--- |
+| **Reasoning** | The reasoning modes available before 1.3 remained available at launch. Meta said `max` reasoning would follow after additional safety testing; the public announcement does not enumerate the other API mode names. |
+| **Coding** | Trained on longer-horizon coding tasks. Meta reports roughly 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2 in internal engineering comparisons. |
+| **Agentic Workflows** | Uses tools to gather context, revises plans, tracks learning across long tasks, asks for clarification, and is trained to confirm consequential actions. |
+| **Multimodal** | Meta's launch demonstrations cover workflows involving documents, spreadsheets, CAD data, audio, and presentations. |
+| **Context Window** | 1M tokens. |
+| **Max Output** | 131,000 tokens. |
+
+### Known Limitations
+
+- `max` reasoning was used for Meta's published Muse Spark 1.3 evaluations but was not available to API users at launch; Meta said it would ship after additional safety testing.
+- Improved self-awareness and safety calibration do not guarantee correct outputs or safe autonomous execution; applications still need approval gates for consequential actions.
+
+## Pricing
+
+| Model | Input (Credits/Token) | Cache Write (Credits/Token) | Cache Read (Credits/Token) | Output (Credits/Token) |
+| :--- | ---: | ---: | ---: | ---: |
+| Muse Spark 1.3 | 1.25 | 1.25 | 0.15 | 4.25 |
diff --git a/docs/llmservice/pricing-and-usage.md b/docs/llmservice/pricing-and-usage.md
index 620754d..9576b85 100644
--- a/docs/llmservice/pricing-and-usage.md
+++ b/docs/llmservice/pricing-and-usage.md
@@ -34,6 +34,7 @@ The table below lists standard reference prices only. For current limited-time o
 | Qwen3.8-Max       |                  2.00 |                        2.00 |                       0.25 |                   6.00 |                        - |
 | Qwen3.7-Max       |                  1.65 |                        1.65 |                       0.33 |                  4.951 |                        - |
 | Qwen3.6-27B       |                  0.19 |                        0.19 |                      0.019 |                   2.99 |                        - |
+| Hy4 Preview       |                 0.834 |                       0.834 |                      0.042 |                  2.501 |                        - |
 | Hy3               |                 0.132 |                       0.132 |                      0.033 |                  0.528 |                        - |
 | MiMo-V2.5-Pro     |                 0.435 |                       0.435 |                     0.0036 |                   0.87 |                        - |
 | MiMo-V2.5         |                  0.14 |                        0.14 |                     0.0028 |                   0.28 |                        - |
@@ -60,6 +61,7 @@ The table below lists standard reference prices only. For current limited-time o
 | GPT-5.4 Nano      |                  0.20 |                        0.20 |                       0.02 |                   1.25 |                   10,000 |
 | GPT-5 Nano        |                  0.05 |                        0.05 |                      0.005 |                   0.40 |                        - |
 | Claude Opus 5     |                  5.00 |                        6.25 |                       0.50 |                  25.00 |                   10,000 |
+| Claude Fable 5.1  |                 10.00 |                       12.50 |                       0.25 |                  50.00 |                        - |
 | Claude Fable 5    |                 10.00 |                       12.50 |                       1.00 |                  50.00 |                   10,000 |
 | Claude Opus 4.8   |                  5.00 |                        6.25 |                       0.50 |                  25.00 |                   10,000 |
 | Claude Opus 4.7   |                  5.00 |                        6.25 |                       0.50 |                  25.00 |                   10,000 |
@@ -69,6 +71,8 @@ The table below lists standard reference prices only. For current limited-time o
 | Claude Sonnet 4.6 |                  3.00 |                        3.75 |                       0.30 |                  15.00 |                   10,000 |
 | Claude Sonnet 4.5 |                  3.00 |                        3.75 |                       0.30 |                  15.00 |                   10,000 |
 | Claude Haiku 4.5  |                  1.00 |                        1.25 |                       0.10 |                   5.00 |                   10,000 |
+| Muse Spark 1.3    |                  1.25 |                        1.25 |                       0.15 |                   4.25 |                        - |
+| Gemini 3.8 Flash  |                  0.75 |                        0.75 |                      0.075 |                   3.75 |                   14,000 |
 | Gemini 3.6 Flash  |                  1.50 |                        1.50 |                       0.15 |                   7.50 |                   14,000 |
 | Gemini 3.5 Flash  |                  1.50 |                        1.50 |                       0.15 |                   9.00 |                   14,000 |
 | Gemini 3.5 Flash-Lite |               0.30 |                        0.30 |                       0.03 |                   2.50 |                   14,000 |
@@ -76,7 +80,7 @@ The table below lists standard reference prices only. For current limited-time o
 | Gemini 3 Flash    |                  0.50 |                        0.50 |                       0.05 |                   3.00 |                   14,000 |
 
 :::caution Main table scope
-The main pricing table shows the currently effective standard reference price for each model. DeepSeek V4 Pro, DeepSeek-V4-Flash, and DeepSeek-V4-Flash-Vision-Exp use time-based pricing, so their rows list both **Idle** and **Busy** rates. Under DeepSeek's updated rule, from 00:00 UTC+8 on August 23, 2026, Saturdays and Sundays are billed at Idle rates all day; the applicable weekday period and final billing are subject to the platform display. DeepSeek-V4-Flash is currently free on B.AI Chat and API, while DeepSeek-V4-Flash-Vision-Exp is currently free for B.AI API use; both rows show standard reference prices. MiMo-V2.5 API usage is currently free, and its B.AI Chat free-access date is August 25, 2026; the exact availability time is subject to actual activation. Its row shows the standard reference price. The `Cache Write` column represents the billing rate when cache writing occurs; it does not imply a unified cache TTL across all models. Cache behavior, retention time, long-context pricing, and extended caching options may vary by model provider. If a model has special caching rules, long-context pricing, 1-hour cache write pricing, or time-based pricing, please refer to the corresponding model detail page.
+The main pricing table shows the currently effective standard reference price for each model. DeepSeek V4 Pro, DeepSeek-V4-Flash, and DeepSeek-V4-Flash-Vision-Exp use time-based pricing, so their rows list both **Idle** and **Busy** rates. DeepSeek API calls use UTC+8: Busy periods are 09:00-12:00 and 14:00-18:00, Monday through Friday; all other times, including weekends, are Idle periods. DeepSeek models used in B.AI Chat are billed at Idle rates. From 17:00 on September 3, 2026 (UTC+8), eligible API usage for DeepSeek-V4-Flash and DeepSeek-V4-Flash-Vision-Exp is billed at 50% of the standard price for the applicable period; both rows continue to show standard reference prices. MiMo-V2.5 API usage is currently free, and its B.AI Chat free-access date is August 25, 2026; the exact availability time is subject to actual activation. Its row shows the standard reference price. The `Cache Write` column represents the billing rate when cache writing occurs; it does not imply a unified cache TTL across all models. Cache behavior, retention time, long-context pricing, and extended caching options may vary by model provider. If a model has special caching rules, long-context pricing, 1-hour cache write pricing, or time-based pricing, please refer to the corresponding model detail page.
 :::
 
 :::info Pricing note
diff --git a/docs/llmservice/promotions-and-pricing-notices.md b/docs/llmservice/promotions-and-pricing-notices.md
index 9793f97..a988921 100644
--- a/docs/llmservice/promotions-and-pricing-notices.md
+++ b/docs/llmservice/promotions-and-pricing-notices.md
@@ -12,31 +12,30 @@ This section only shows currently active promotions: free offers appear first, f
 
 <ActivityCard
   variant="free"
-  title="DeepSeek-V4-Flash"
+  title="GLM-5.3-Flash"
   status="Free Offer"
-  detail="0 Credits"
+  detail="API Free · Chat on Launch"
 >
-Offer starts August 17, 2026.
-
-This offer applies to B.AI Chat and API:
+This offer covers B.AI API and Chat:
 
-* **Chat:** DeepSeek-V4-Flash usage is billed at `0 Credits` during the offer.
-* **API:** Usage is billed at `0 Credits` during the offer. No per-request, input, cache write, cache read, or output token fees apply.
+* **API:** GLM-5.3-Flash API usage is currently billed at `0 Credits`. No input, cache write, cache read, or output token fees apply.
+* **Chat:** Free access begins when GLM-5.3-Flash becomes available in B.AI Chat. The availability time is subject to the actual model listing. Once available, Chat usage is billed at `0 Credits`.
 
-After the offer ends, DeepSeek-V4-Flash will return to standard pricing. See the [model details](./models/deepseek-v4-flash.md).
+After the offer ends, GLM-5.3-Flash will return to the prices shown in the [model details](./models/glm-5-3-flash.md).
 </ActivityCard>
 
 <ActivityCard
   variant="free"
-  title="DeepSeek-V4-Flash-Vision-Exp"
-  status="API Free"
-  detail="0 Credits"
+  title="Qwen3.8-Flash"
+  status="Free Offer"
+  detail="API Free · Chat on Launch"
 >
-This offer is currently active and applies only to `deepseek-v4-flash-vision-exp` requests made through the B.AI API.
+Free access is available in phases across B.AI API and Chat:
 
-During the offer, API usage is billed at `0 Credits`. No per-request, input, cache write, cache read, or output token fees apply.
+* **API:** Qwen3.8-Flash API usage is currently billed at `0 Credits`. No input, cache write, cache read, or output token fees apply.
+* **Chat:** Free access begins when Qwen3.8-Flash becomes available in B.AI Chat. The availability date is subject to the actual model listing. Once available, Chat usage is billed at `0 Credits`.
 
-After the offer ends, the model will return to standard pricing. See [Pricing and Usage](./pricing-and-usage.md) for the standard reference price.
+After the offer ends, Qwen3.8-Flash will return to the prices shown in the [model details](./models/qwen3-8-flash.md).
 </ActivityCard>
 
 <ActivityCard
@@ -70,31 +69,29 @@ After the offer ends, MiMo-V2.5 will return to standard pricing. See the [model
 </ActivityCard>
 
 <ActivityCard
-  variant="free"
-  title="GLM-5.3-Flash"
-  status="Free Offer"
-  detail="API Free · Chat on Launch"
+  variant="discount"
+  title="DeepSeek-V4-Flash"
+  status="Limited-Time API Discount"
+  detail="50% of Standard Price"
 >
-This offer covers B.AI API and Chat:
+The DeepSeek-V4-Flash 50% offer takes effect at 17:00 on September 3, 2026 (UTC+8).
 
-* **API:** GLM-5.3-Flash API usage is currently billed at `0 Credits`. No input, cache write, cache read, or output token fees apply.
-* **Chat:** Free access begins when GLM-5.3-Flash becomes available in B.AI Chat. The availability time is subject to the actual model listing. Once available, Chat usage is billed at `0 Credits`.
+From the effective time, eligible DeepSeek-V4-Flash API usage is billed at 50% of the standard price for the applicable period. The discounted price changes in step with DeepSeek's Idle and Busy pricing periods and remains at 50% in either period.
 
-After the offer ends, GLM-5.3-Flash will return to the prices shown in the [model details](./models/glm-5-3-flash.md).
+The [Pricing and Usage](./pricing-and-usage.md) table and [model details](./models/deepseek-v4-flash.md) continue to show standard reference prices. Actual settlement and final billing are subject to the platform display.
 </ActivityCard>
 
 <ActivityCard
-  variant="free"
-  title="Qwen3.8-Flash"
-  status="Free Offer"
-  detail="API Free · Chat on Launch"
+  variant="discount"
+  title="DeepSeek-V4-Flash-Vision-Exp"
+  status="Limited-Time API Discount"
+  detail="50% of Standard Price"
 >
-Free access is available in phases across B.AI API and Chat:
+The DeepSeek-V4-Flash-Vision-Exp 50% offer takes effect at 17:00 on September 3, 2026 (UTC+8).
 
-* **API:** Qwen3.8-Flash API usage is currently billed at `0 Credits`. No input, cache write, cache read, or output token fees apply.
-* **Chat:** Free access begins when Qwen3.8-Flash becomes available in B.AI Chat. The availability date is subject to the actual model listing. Once available, Chat usage is billed at `0 Credits`.
+From the effective time, eligible `deepseek-v4-flash-vision-exp` API usage is billed at 50% of the standard price for the applicable period. The discounted price changes in step with DeepSeek's Idle and Busy pricing periods and remains at 50% in either period.
 
-After the offer ends, Qwen3.8-Flash will return to the prices shown in the [model details](./models/qwen3-8-flash.md).
+The [Pricing and Usage](./pricing-and-usage.md) table and [model details](./models/deepseek-v4-flash-vision-exp.md) continue to show standard reference prices. Actual settlement and final billing are subject to the platform display.
 </ActivityCard>
 
 <ActivityCard
@@ -131,7 +128,7 @@ For a limited time, eligible requests are billed at 90% of the standard referenc
   status="Time-Based Pricing"
   detail="Idle / Busy Rates"
 >
-DeepSeek-V4-Pro uses Idle and Busy pricing periods. The applicable period and final billing are subject to the platform display.
+DeepSeek API calls use time-based pricing in UTC+8. Busy periods are 09:00-12:00 and 14:00-18:00, Monday through Friday; all other times, including weekends, are Idle periods. DeepSeek models used in B.AI Chat are billed at Idle rates.
 
-DeepSeek-V4-Flash also uses Idle and Busy pricing periods, but is currently free across B.AI Chat and API under its limited-time offer. See the [DeepSeek-V4-Pro](./models/deepseek-v4-pro.md) and [DeepSeek-V4-Flash](./models/deepseek-v4-flash.md) model details for standard pricing information. Final billing is subject to the platform display.
+DeepSeek-V4-Pro, DeepSeek-V4-Flash, and DeepSeek-V4-Flash-Vision-Exp use these Idle and Busy periods. From 17:00 on September 3, 2026 (UTC+8), eligible DeepSeek-V4-Flash and DeepSeek-V4-Flash-Vision-Exp API usage is billed at 50% of the standard price for the applicable period. See the [DeepSeek-V4-Pro](./models/deepseek-v4-pro.md) and [DeepSeek-V4-Flash](./models/deepseek-v4-flash.md) model details for standard pricing information. Final billing is subject to the platform display.
 </ActivityCard>
diff --git a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/api/API.md b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/api/API.md
index 07179a7..1d999ac 100644
--- a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/api/API.md
+++ b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/api/API.md
@@ -138,7 +138,9 @@ Responses API 与 Chat Completions 的请求结构不同：
 
 ### 模型与端点兼容性
 
-Responses 端点接受已启用该协议的模型。模型与端点不兼容时，服务器返回 HTTP `400` 及对应错误信息。
+Responses 端点支持 B.AI 提供的 GPT 与 DeepSeek 模型系列。可以通过 `GET /v1/models` 获取当前 API 凭证可用的模型 ID。不同模型支持的参数和工具可能不同；模型与端点不兼容时，服务器返回 HTTP `400` 及对应错误信息。
+
+DeepSeek 模型不支持联网搜索。通过 Responses API 调用 DeepSeek 模型时，请勿在请求中传入联网搜索工具；在 Codex 中使用 DeepSeek 模型时，需要将 Codex 顶层配置设置为 `web_search = "disabled"`。
 
 ### 请求体
 
@@ -669,7 +671,7 @@ curl https://api.b.ai/v1/messages \
 
 ## Codex CLI 接入
 
-B.AI Responses API 可以作为 Codex 的自定义模型提供商使用。以下配置适用于支持自定义模型提供商的 Codex 版本。
+B.AI Responses API 可以作为 Codex 的自定义模型提供商，并支持相应的 GPT 与 DeepSeek 模型系列。以下配置适用于支持自定义模型提供商的 Codex 版本。
 
 ### 1. 设置 API Key
 
@@ -699,6 +701,20 @@ wire_api = "responses"
 requires_openai_auth = false
 ```
 
+:::caution 使用 DeepSeek 前必须关闭 Codex 联网搜索
+Codex 默认可能启用内置联网搜索工具，但 DeepSeek 模型不支持该工具。如果没有关闭，请求可能返回“不支持 Web Search 工具”的错误。
+
+打开 `~/.codex/config.toml`，增加顶层配置 `web_search = "disabled"`：
+
+```toml
+model = "your-model-id"
+model_provider = "bai"
+web_search = "disabled"
+```
+
+请将 `web_search` 放在 `[model_providers.bai]` 配置块上方和配置块之外。保存文件后，完全退出并重新启动 Codex。该设置是使用 DeepSeek 时的必要配置；切换到需要联网搜索的模型配置时，应删除或调整此项。
+:::
+
 如果该文件已有内容，把 `[model_providers.bai]` 整块追加进去，并将顶层的 `model` 与 `model_provider` 改成上面的取值。各配置项的完整说明见文末的 Codex 文档。
 
 保存后，在已经设置 `BAI_API_KEY` 的终端中启动 Codex：
@@ -722,6 +738,7 @@ model = "your-model-id"
 | 返回 `401` | 检查 API Key 是否有效，以及是否误用了其他环境的 Key。 |
 | 返回 `403` | 检查账户状态与模型权限。 |
 | 返回模型不支持 | 确认模型 ID 拼写正确，并已为所配置的端点启用。 |
+| DeepSeek 请求提示不支持联网搜索工具 | 确认顶层配置包含 `web_search = "disabled"`，然后重启 Codex。 |
 
 ---
 
@@ -788,6 +805,7 @@ model = "your-model-id"
 | 场景 | 状态码 | 处理方式 |
 |---|---:|---|
 | 模型与端点不兼容 | `400` | 选择已为该端点启用的模型，或改用其他端点。 |
+| DeepSeek 请求包含联网搜索工具 | `400` | 移除联网搜索工具；在 Codex 中使用时，将顶层配置设为 `web_search = "disabled"`。 |
 | 使用 `max_tokens` 或 `max_completion_tokens` | `400` | 改用 `max_output_tokens`。 |
 | `max_output_tokens` 超出模型允许范围 | `400` | 按错误信息给出的范围调整取值。 |
 | 请求使用了不可用的工具 | `400` | 移除该工具，或选择兼容的模型配置。 |
diff --git a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/claude-fable-5-1.md b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/claude-fable-5-1.md
new file mode 100644
index 0000000..7e3bb98
--- /dev/null
+++ b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/claude-fable-5-1.md
@@ -0,0 +1,56 @@
+# Claude Fable 5.1
+
+## 概述
+
+Claude Fable 5.1 是 Anthropic 于 2026 年 9 月 1 日发布的 Claude 5 系列 Mythos 级通用模型，面向高难度推理和长期 Agent 工作流。该模型与限量开放的 Claude Mythos 5.1 采用相同的底层模型。在 B.AI 上使用时，模型 ID 为 `claude-fable-5-1`。
+
+## 核心特性
+
+* **长周期 Agent 能力：** 在启用生产环境安全机制的情况下，Anthropic 公布其 Terminal-Bench-Science 0.1 得分为 52.6%、Terminal-Bench 4.0 得分为 55.8%、AutomationBench 得分为 31.4%，CursorBench 3.2.0 得分为 73.4%。
+* **始终启用自适应思考：** 自适应思考无法关闭。`effort` 支持 `low`、`medium`、`high`、`xhigh` 和 `max`，Claude API 默认使用 `high`。
+* **1M 上下文与 128K 输出：** 支持 1,000,000 Token 上下文窗口和最高 128,000 Tokens 输出，完整上下文窗口均按标准 Token 价格计费。
+* **面向 Agent 的 API 控制：** 支持自动工具调用、严格工具 Schema、结构化输出、保留思考内容、按消息设置 effort、按轮次设置系统消息、可读进度更新和内容来源标记。部分控制能力处于 Beta 阶段。
+* **更低的缓存读取价格：** 缓存命中和刷新价格为 `0.25 Credits/Token`，是 Claude Fable 5 缓存读取价格的四分之一；基础输入和输出价格保持不变。
+
+## 适用场景
+
+* **长期软件工程任务：** 适用于代码仓库级实现、调试、代码审查、性能优化，以及跨越多个工具和检查点的自主任务。
+* **多步骤研究与分析：** 适用于需要搜索、证据综合、迭代验证假设，以及从失败步骤中恢复的研究工作流。
+* **企业知识工作：** 适用于复杂文档、电子表格、演示文稿、财务分析和跨文件工作流，可充分利用大上下文窗口和视觉能力。
+* **浏览器与计算机操作 Agent：** 适用于跨应用操作、长时间保持任务状态，以及在工具调用之间提供进度更新的任务。
+
+## 能力与限制
+
+| 能力 | 说明 |
+| :--- | :--- |
+| **推理** | 始终启用自适应思考，并支持五档 effort。Anthropic 公布其 Humanity's Last Exam 无工具得分为 60.9%，使用工具时为 65.0%。 |
+| **创意写作** | 支持撰写、编辑、研究综合和专业内容生成。 |
+| **多模态** | 支持文本和图像输入、文本输出，可分析图表、示意图、表格、文件和 PDF。 |
+| **上下文窗口** | 1M Tokens。 |
+| **最大输出** | 128,000 Tokens。 |
+| **工具调用** | 支持自动函数调用、严格工具调用、结构化输出、浏览器与计算机操作工作流，以及保留思考内容。不支持强制选择 `any` 或指定工具。 |
+| **多语言** | Anthropic 将多语言支持列为当前 Claude 模型的通用能力。 |
+| **知识截止时间** | 可靠知识截止时间及训练数据截止时间：2026 年 6 月。 |
+
+### 已知限制
+
+* 不支持强制工具选择。将 `tool_choice` 设置为 `any` 或指定工具会返回 400 错误；集成时应使用 `auto` 或 `none`，并在 Prompt 中描述工具调用要求。
+* 思考块仅支持向前兼容：Claude Fable 5.1 可以读取早期模型的思考块，但早期模型无法读取 Claude Fable 5.1 的思考块。
+* 在 Fable 5.1 思考块之后修改系统提示词、工具定义或更早的消息，可能导致该思考块失效。长期运行的集成应将会话历史视为仅追加内容，或使用文档中说明的思考块丢弃控制。
+* 该模型响应速度相对较慢，基础输入和输出 Token 价格是 Claude Opus 5 的两倍。
+* Fable 安全机制可能拒绝部分网络安全和生物学请求，或将其路由至其他 Claude 模型。Fable 5.1 默认要求保留数据 30 天；零数据保留需要获得 Anthropic 的明确授权。
+* Claude Fable 5.1 不支持 Priority Tier。
+
+## 定价
+
+| 模型 | 输入（Credits/Token） | 5 分钟缓存写入（Credits/Token） | 1 小时缓存写入（Credits/Token） | 缓存读取（Credits/Token） | 输出（Credits/Token） | 网页搜索（Credits/次） |
+| :--- | --------------------: | -----------------------------: | -----------------------------: | ------------------------: | --------------------: | ---------------------: |
+| **Claude Fable 5.1** | `10.00` | `12.50` | `20.00` | `0.25` | `50.00` | `-` |
+
+:::info 缓存说明
+Claude Fable 5.1 的 5 分钟缓存写入按输入价格的 1.25 倍计费，1 小时缓存写入按输入价格的 2 倍计费；缓存读取按输入价格的 0.025 倍计费。
+:::
+
+:::info 价格说明
+文档价格为 B.AI 平台模型标准参考价，仅供基础计费说明使用。B.AI 可能会通过充值赠送及账户权益等方式，为用户提供更低的实际使用成本。具体价格、赠送积分及账户权益请以平台页面展示及最终账单为准。
+:::
diff --git a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/deepseek-v4-flash-vision-exp.md b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/deepseek-v4-flash-vision-exp.md
new file mode 100644
index 0000000..d78bb17
--- /dev/null
+++ b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/deepseek-v4-flash-vision-exp.md
@@ -0,0 +1,67 @@
+import ActivityCard from '@site/src/components/ActivityCard';
+
+# DeepSeek-V4-Flash-Vision-Exp
+
+<ActivityCard
+  variant="discount"
+  title="DeepSeek-V4-Flash-Vision-Exp"
+  status="API 限时折扣"
+  detail="标准价 5 折"
+>
+DeepSeek-V4-Flash-Vision-Exp 5 折活动于 2026 年 9 月 3 日 17:00（UTC+8）正式生效。
+
+自生效时间起，符合条件的 `deepseek-v4-flash-vision-exp` API 使用按对应闲时或忙时标准价的 50% 结算。活动价格随 DeepSeek 分时计费时段同步变化，各时段均保持 5 折。
+
+下方价格表继续展示标准参考价，实际结算价格及最终账单以平台页面展示为准。
+</ActivityCard>
+
+## 概述
+
+DeepSeek-V4-Flash-Vision-Exp 是 DeepSeek-V4 系列的首款实验性多模态模型，于 2026 年 8 月 21 日发布。该模型在 DeepSeek-V4-Flash 基础上增加视觉模块并继续进行图像理解训练，同时在 DeepSeek 的评测中保持了相近的纯文本智能体能力。
+
+## 主要特性
+
+- **原生图像理解**：支持以 JPEG、PNG、GIF 和 WebP 格式交错输入文本与图像，可用于读取截图、分析图表、视觉问答及其他基于图像的任务。
+- **多模态智能体评测**：DeepSeek 公布的结果为 ApexBench（Pass@1）36.5、Chartography 64.3 和 ZeroBench（Pass@5）35.0，显示其视觉智能体能力较纯文本 DeepSeek-V4-Flash-0731 基线有所提升。
+- **延续文本智能体能力**：DeepSeek 表示其纯文本能力与 DeepSeek-V4-Flash 相近，公布的成绩包括 Terminal-Bench 2.1 83.9、DeepSWE 59.3 和 Toolathlon-Verified 75.9。
+- **长上下文推理**：托管 API 提供 1M Token 上下文窗口和最高 384K Tokens 输出，支持可选思考模式以及 `low`、`high` 或 `max` 推理强度。
+- **面向智能体的 API 支持**：支持工具调用、JSON 输出、上下文缓存和 Chat Prefix Completion，同时支持 OpenAI 兼容的 Chat Completions、Responses 接口及 Anthropic 兼容的 Messages 接口。
+- **开放权重参考版本**：官方 MIT License 仓库提供 305B 参数检查点，并包含视觉编码器与对齐器、DFlash Attention、混合专家层、Hyper-Connections 和 DSpark 前向路径的参考代码。
+
+## 适用场景
+
+- **视觉软件智能体**：结合工具检查截图、渲染后的界面、仪表盘或画布状态，并据此修改代码或其他内容。
+- **文档和图表分析**：读取截图中的文本、理解图表，并根据视觉化的业务或技术资料回答问题。
+- **多模态工具工作流**：通过 Responses API 处理工具返回的图像，将视觉证据与多步推理和函数调用结合。
+- **大批量视觉审查**：在单次请求中处理多张图像，对截图、页面或帧集合执行一致的分析。
+- **开放权重评测**：在能够承担大型检查点和参考运行环境要求的受控自托管环境中，测试 DeepSeek 实验性 V4 视觉架构。
+
+## 能力与限制
+
+| 能力 | 说明 |
+| :--- | :--- |
+| **推理** | 支持思考和非思考模式，思考模式默认使用 `high` 强度。主要推理强度为 `low`、`high` 和 `max`，兼容的 `medium` 与 `xhigh` 请求会映射为 `high`。 |
+| **编程与智能体** | DeepSeek 公布的结果为 Terminal-Bench 2.1 83.9、DeepSWE 59.3 和 Toolathlon-Verified 75.9。这些厂商评测使用 DeepSeek Harness minimal 模式、`max` 强度、`temperature = 1.0` 和 `top_p = 0.95`。 |
+| **工具调用** | 在思考和非思考模式下均支持函数调用，同时支持 JSON 输出、上下文缓存及 Responses API 中包含图像的工具输出。 |
+| **多模态** | 支持文本与图像输入，输出为文本。每张图像会在推理前调整尺寸，最多占用 384 个输入 Tokens；在尺寸限制范围内，API 每次请求最多接受 600 张图像。 |
+| **上下文窗口** | 1M Tokens。 |
+| **最大输出** | 384K Tokens。 |
+
+### 已知限制
+
+- 图像可放置的位置取决于接口：Chat Completions 接受 `user` 消息中的图像；Responses API 还接受 `developer` 消息和受支持工具输出中的图像。`system` 或 `assistant` 消息中的图像会被拒绝。
+- 已记录的多模态接口支持图像，但未说明支持音频、视频输入、图像生成或其他非文本输出。
+- 不支持 FIM Completion，Chat Prefix Completion 仍为 Beta 功能。
+- 在思考模式下，`temperature`、`top_p`、`presence_penalty` 和 `frequency_penalty` 不生效。包含工具调用的多轮请求需要在后续请求中完整回传此前的 `reasoning_content`。
+- DeepSeek 未公布该模型专属的知识截止时间或完整语言支持列表。
+
+## 价格
+
+| 计费时段 | 输入（Credits/Token） | 缓存写入（Credits/Token） | 缓存读取（Credits/Token） | 输出（Credits/Token） |
+| :--- | ---: | ---: | ---: | ---: |
+| **闲时** | 0.22 | 0.22 | 0.0073 | 0.66 |
+| **忙时** | 0.44 | 0.44 | 0.0147 | 1.32 |
+
+API 调用按北京时间（UTC+8）实行分时计费：忙时为周一至周五 09:00-12:00 和 14:00-18:00；其余时间（包括周六、周日全天）为闲时。
+
+图像会根据尺寸换算为输入 Tokens，并与文本输入一同计费；调整尺寸后，每张图像最多占用 384 Tokens。
diff --git a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/deepseek-v4-flash.md b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/deepseek-v4-flash.md
index c5efd6c..e6c5574 100644
--- a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/deepseek-v4-flash.md
+++ b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/deepseek-v4-flash.md
@@ -7,17 +7,16 @@ import ActivityCard from '@site/src/components/ActivityCard';
 DeepSeek-V4-Flash 是 DeepSeek 于 2026 年 4 月 24 日与 V4-Pro 同步发布的高效率开源大语言模型，采用 MIT License。该模型总参数量为 284B，但每次前向仅激活 13B 参数，以仅为 V4-Pro 约 1/3.1 的成本提供接近旗舰模型的性能，是当前极具性价比的模型之一。
 
 <ActivityCard
-  variant="free"
+  variant="discount"
   title="DeepSeek-V4-Flash"
-  status="免费活动"
-  detail="0 Credits"
+  status="API 限时折扣"
+  detail="标准价 5 折"
 >
-活动开始时间：2026 年 8 月 17 日，适用于 B.AI Chat 和 API 中的 DeepSeek-V4-Flash。
+DeepSeek-V4-Flash 5 折活动于 2026 年 9 月 3 日 17:00（UTC+8）正式生效。
 
-* **Chat：** 活动期间使用按 `0 Credits` 结算。
-* **API：** 活动期间使用按 `0 Credits` 结算，不收取单次请求、输入、缓存写入、缓存读取或输出的费用。
+自生效时间起，符合条件的 DeepSeek-V4-Flash API 使用按对应闲时或忙时标准价的 50% 结算。活动价格随 DeepSeek 闲时与忙时计费时段同步变化，各时段均保持 5 折。
 
-活动结束后，模型将恢复标准价格。活动结束时间、适用规则、实际结算结果及最终账单以平台页面展示为准。
+下方价格表继续展示标准参考价，实际结算价格及最终账单以平台页面展示为准。
 </ActivityCard>
 
 ## 核心特性
@@ -60,5 +59,5 @@ DeepSeek-V4-Flash 是 DeepSeek 于 2026 年 4 月 24 日与 V4-Pro 同步发布
 | **忙时** | `0.44` | `0.44` | `0.0147` | `1.32` | `-` | 缓存写入：输入价的 `1x`；缓存读取：输入价的 `0.0333x` |
 
 :::info 价格说明
-本表展示 DeepSeek-V4-Flash 的分时标准参考价。根据 DeepSeek 官方最新规则，自 2026 年 8 月 23 日 00:00（UTC+8）起，周六、周日全天按闲时价格计费。当前限时活动期间，B.AI Chat 和 API 的所有使用均按 `0 Credits` 结算。活动结束后，工作日具体适用闲时或忙时时段、实际结算价格及最终账单以平台页面展示为准。B.AI 可能会通过充值赠送及账户权益等方式，为用户提供更低的实际使用成本。
+本表展示 DeepSeek-V4-Flash 的分时标准参考价。API 调用按北京时间（UTC+8）实行分时计费：忙时为周一至周五 09:00-12:00 和 14:00-18:00；其余时间（包括周六、周日全天）为闲时。B.AI Chat 中的 DeepSeek-V4-Flash 按闲时价格计费。自 2026 年 9 月 3 日 17:00（UTC+8）起，符合条件的 API 使用按调用发生时对应闲时或忙时标准价的 50% 结算。实际结算价格及最终账单以平台页面展示为准。B.AI 可能会通过充值赠送及账户权益等方式，为用户提供更低的实际使用成本。
 :::
diff --git a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/deepseek-v4-pro.md b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/deepseek-v4-pro.md
index b0d1bf0..6637fe6 100644
--- a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/deepseek-v4-pro.md
+++ b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/deepseek-v4-pro.md
@@ -44,5 +44,5 @@ DeepSeek-V4-Pro 是 DeepSeek 于 2026 年 4 月 24 日基于 MIT License 发布
 | **忙时** | `1.32` | `1.32` | `0.044` | `3.96` | `-` | 缓存写入：输入价的 `1x`；缓存读取：输入价的 `0.0333x` |
 
 :::info 价格说明
-DeepSeek V4 Pro 采用分时标准参考价。根据 DeepSeek 官方最新规则，自 2026 年 8 月 23 日 00:00（UTC+8）起，周六、周日全天按闲时价格计费。工作日具体适用闲时或忙时时段、实际结算价格及最终账单以平台页面展示为准。B.AI 可能会通过充值赠送及账户权益等方式，为用户提供更低的实际使用成本。
+DeepSeek V4 Pro 采用分时标准参考价。API 调用按北京时间（UTC+8）实行分时计费：忙时为周一至周五 09:00-12:00 和 14:00-18:00；其余时间（包括周六、周日全天）为闲时。B.AI Chat 中的 DeepSeek V4 Pro 按闲时价格计费。实际结算价格及最终账单以平台页面展示为准。B.AI 可能会通过充值赠送及账户权益等方式，为用户提供更低的实际使用成本。
 :::
diff --git a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/gemini-3-8-flash.md b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/gemini-3-8-flash.md
new file mode 100644
index 0000000..7f6395d
--- /dev/null
+++ b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/gemini-3-8-flash.md
@@ -0,0 +1,47 @@
+# Gemini 3.8 Flash
+
+## 概述
+
+Gemini 3.8 Flash 是 Google 于 2026 年 9 月 2 日正式发布的 Gemini 3 系列模型，面向长周期软件工程、自主智能体和复杂知识工作流，提供 1M Token 输入窗口及可配置的推理强度。
+
+## 主要特性
+
+- **长周期智能体执行**：适合需要多步规划、反复调用工具和持续验证的软件工程及企业工作流。
+- **可配置思考强度**：支持 `low`、`medium` 和 `high` 三档思考级别，默认为 `medium`，不支持 `minimal`。
+- **原生多模态输入**：支持文本、图像、视频、音频和 PDF 输入，输入上限为 1,048,576 Tokens，最多输出 65,536 个文本 Tokens。
+- **丰富的工具能力**：支持函数调用、结构化输出、代码执行、File Search、URL Context、搜索与地图信息增强，以及预览阶段的 Computer Use。
+
+## 适用场景
+
+- **智能体软件工程**：适用于代码仓库分析、多文件开发、调试和迁移等需要持续推理与工具调用的任务。
+- **复杂知识工作**：适用于基于大型文档集、结构化数据和专业资料的长篇分析与报告生成。
+- **多模态分析**：结合文本指令理解文档、截图、图表、视频和音频内容。
+- **工具编排应用**：构建结合自定义函数、代码执行、搜索、URL 获取、文件搜索或计算机操作的智能体。
+- **异步批量处理**：适用于可使用 Batch 或 Flex 推理的分类、抽取、评测等非实时工作负载。
+
+## 能力与限制
+
+| 能力 | 说明 |
+| :--- | :--- |
+| **推理** | 支持 `low`、`medium` 和 `high` 思考级别，默认为 `medium`。提高思考强度可能改善复杂多步任务的效果，同时也可能增加 Token 消耗和延迟。 |
+| **编程** | 针对长周期软件工程和智能体编程工作流优化。Google 发布材料显示其相较 Gemini 3.7 Flash 有所提升，但厂商评测结果不代表生产环境表现保证。 |
+| **多模态** | 支持文本、图像、视频、音频和 PDF 输入，输出为文本。 |
+| **上下文窗口** | 1,048,576 个输入 Tokens。 |
+| **最大输出** | 65,536 Tokens。 |
+| **工具调用** | 支持函数调用、结构化输出、代码执行、File Search、URL Context、搜索与地图信息增强，以及 Computer Use（预览）。 |
+| **知识截止时间** | 整体知识截止至 2026 年 3 月；Google 说明部分领域的知识可能仅覆盖至 2025 年 1 月。处理较新或时效性信息时建议启用信息增强。 |
+
+### 已知限制
+
+- 模型可能产生不准确内容，Google 也提示偶尔可能出现响应较慢或超时。
+- 复杂任务和较高思考级别可能比以往的 Flash 工作流消耗更多推理及输出 Tokens，应用应根据延迟和成本目标调整思考强度。
+- 不支持 `minimal` 思考级别。`temperature`、`top_p` 和 `top_k` 等旧版采样参数会被忽略，也不支持预填充模型回复。
+- Computer Use 仍处于预览阶段，模型不支持原生图像生成或原生音频生成。
+
+## 价格
+
+| 模型 | 输入（Credits/Token） | 缓存写入（Credits/Token） | 缓存读取（Credits/Token） | 输出（Credits/Token） | 网页搜索（Credits/次） |
+| :--- | ---: | ---: | ---: | ---: | ---: |
+| Gemini 3.8 Flash | 0.75 | 0.75 | 0.075 | 3.75 | 14,000 |
+
+显式缓存存储费用在 2026 年 12 月 31 日前为每小时 0.50 Credits/Token，自 2027 年 1 月 1 日起为每小时 1.00 Credits/Token。
diff --git a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/hy4-preview.md b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/hy4-preview.md
new file mode 100644
index 0000000..b2d312d
--- /dev/null
+++ b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/hy4-preview.md
@@ -0,0 +1,53 @@
+# Hy4 Preview
+
+## 概述
+
+Hy4 Preview 是腾讯混元团队于 2026 年 8 月 28 日发布的开源权重混合专家（MoE）语言模型，也是 Hy4 系列的早期版本。模型主干总参数量为 7700 亿，每个 Token 激活 490 亿参数，并支持 1M Token 上下文窗口。在 B.AI 上使用时，模型 ID 为 `hy4-preview`。
+
+## 核心特性
+
+* **770B 稀疏 MoE 架构：** 模型包含 78 层主干网络；每个 MoE 层包含 256 个路由专家和 1 个共享专家，每个 Token 激活 8 个路由专家及共享专家。独立的原生 MTP 层另含 100 亿总参数和 7 亿激活参数，用于推测解码。
+* **长上下文 API 限制：** 支持 1M Token 上下文窗口，最大输入 960K Tokens，最大输出 64K Tokens。
+* **保留深度推理：** 默认启用思考。API 会返回 `reasoning_content`；在多轮工具调用中，应随完整的 Assistant 消息一并回传。
+* **Agent API 支持：** 支持结构化输出、函数调用、缓存输入，以及 OpenAI Chat Completions、OpenAI Responses 和 Anthropic Messages 接口。
+* **面向生产力任务的验证：** 在腾讯内部盲测中，163 位专家对 203 项工程任务的输出进行评分。Hy4 Preview 平均得分为 2.99/4.00，GLM-5.3 为 2.92，Kimi K3 为 2.94。
+
+## 适用场景
+
+* **软件工程 Agent：** 适用于大型代码仓库中的长期规划、实现、调试、验证、前端开发和多步骤工具调用。
+* **办公与数据工作流：** 适用于跨文档分析、财务建模、数据分析，以及文档、电子表格和演示文稿的生成。
+* **游戏开发：** 适用于可玩原型生成、游戏引擎工作流和多轮迭代优化。
+* **科学研究：** 适用于 AI 研究、分子动力学、凝聚态物理和数学等领域的工具辅助分析与问题求解。
+* **私有化或定制部署：** 适用于具备相应计算资源，并需要本地推理、量化或特定任务微调的开源权重部署场景。
+
+## 能力与限制
+
+| 能力 | 说明 |
+| :--- | :--- |
+| **推理** | 支持保留思考内容的深度推理。文档提供 `low` 和 `high` 两档推理强度；在 `hy4-preview` 工具工作流中默认为 `high`。 |
+| **创意写作** | 支持通用文本生成和中文内容创作。 |
+| **多模态** | 文本输入、文本输出。 |
+| **响应速度** | 腾讯未公布有保证的生成速率。官方发布资料显示，借助模型与推理系统协同优化，吞吐量相对腾讯内部基线提升 31.8%。 |
+| **上下文窗口** | 1M Tokens。 |
+| **最大输入** | 960,000 Tokens。 |
+| **最大输出** | 64,000 Tokens。 |
+| **工具调用** | 支持函数调用、结构化输出、自动工具选择、缓存输入，以及在工具调用之间保留推理内容。 |
+| **多语言** | 腾讯文档明确支持中文和英文使用。 |
+
+### 已知限制
+
+* 腾讯将该版本定位为 Hy4 的早期版本，而非最终版 Hy4 模型。
+* 官方模型卡指出，Hy4 Preview 在处理复杂任务时可能进行超过必要程度的推理，并可能出现过度验证。
+* 多轮工具集成应保留并回传完整的 Assistant 消息，包括 `reasoning_content`；省略该字段可能影响后续调用的连续性。
+* Hy4 Preview 不支持原生图像、音频或视频理解。
+* 腾讯尚未公布该模型的具体知识截止时间、有保证的 API 吞吐量或完整语言支持列表。
+
+## 定价
+
+| 模型 | 输入（Credits/Token） | 缓存写入（Credits/Token） | 缓存读取（Credits/Token） | 输出（Credits/Token） | 网页搜索（Credits/次） |
+| :--- | --------------------: | ------------------------: | ------------------------: | --------------------: | ---------------------: |
+| **Hy4 Preview** | `0.834` | `0.834` | `0.042` | `2.501` | `-` |
+
+:::info 价格说明
+文档价格为 B.AI 平台模型标准参考价，仅供基础计费说明使用。B.AI 可能会通过充值赠送及账户权益等方式，为用户提供更低的实际使用成本。具体价格、赠送积分及账户权益请以平台页面展示及最终账单为准。
+:::
diff --git a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/muse-spark-1-3.md b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/muse-spark-1-3.md
new file mode 100644
index 0000000..3f41215
--- /dev/null
+++ b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/muse-spark-1-3.md
@@ -0,0 +1,42 @@
+# Muse Spark 1.3
+
+## 概述
+
+Muse Spark 1.3 是 Meta Superintelligence Labs 于 2026 年 9 月 2 日发布的专有智能体与编程模型。Meta 通过 Muse Code 和 Meta Model API 提供该模型，并将本次更新定位于长周期任务、更强的指令保持、多任务处理以及更高效的编程智能体执行。
+
+## 主要特性
+
+- **长周期智能体工作**：能够借助工具积累上下文、识别计划中的缺口、保留已获得的信息，并在长时间工作流中完成开放式目标。
+- **协作式控制**：当指令存在歧义时会主动澄清，遇到阻塞时请求帮助，并在执行可能产生重要影响的操作前进行确认。
+- **多工作流跟踪**：在长期、被打断或混合用途的会话中，能够更准确地将新提示对应到相关任务，同时保留此前要求。
+- **编程效率**：根据 Meta 工程师进行的对比，模型相较 Muse Spark 1.2 减少约 20% 的工具调用和 25% 的 Token 消耗，同时生成更简洁的代码。
+- **改进安全判断**：Meta 表示该模型对提示词注入的抵抗能力更强，并能在智能体工作流中更谨慎地判断不可逆操作。
+
+## 适用场景
+
+- **长时间运行的编程智能体**：适用于需要反复调用工具和持续遵循指令的功能开发、调试及代码库分析等工程任务。
+- **复杂交付工作流**：基于多个文件及可能存在冲突的来源制作报告、演示文稿、分析或其他交付内容。
+- **研究与自动化智能体**：适用于需要自主收集上下文、调整计划，并结合浏览或业务自动化工具的工作流。
+- **多任务协作**：适用于用户会打断、调整或恢复多个相关工作流的长期助手会话。
+
+## 能力与限制
+
+| 能力 | 说明 |
+| :--- | :--- |
+| **推理** | 1.3 发布时保留了此前可用的推理模式。Meta 表示 `max` 推理将在完成进一步安全测试后提供，公开公告未列出其他 API 模式名称。 |
+| **编程** | 针对长周期编程任务训练。Meta 的内部工程对比显示，其工具调用次数比 Muse Spark 1.2 约少 20%，Token 消耗约少 25%。 |
+| **智能体工作流** | 使用工具收集上下文、调整计划、记录长任务中的学习结果、主动澄清问题，并在执行重要操作前进行确认。 |
+| **多模态** | Meta 的发布演示涵盖涉及文档、电子表格、CAD 数据、音频和演示文稿的工作流。 |
+| **上下文窗口** | 1M Tokens。 |
+| **最大输出** | 131,000 Tokens。 |
+
+### 已知限制
+
+- Meta 发布的 Muse Spark 1.3 评测使用了 `max` 推理，但该模式在发布时尚未向 API 用户开放；Meta 表示将在完成额外安全测试后提供。
+- 改进的自我判断和安全校准不能保证输出始终正确，也不能保证自主执行始终安全；涉及重要后果的操作仍应设置人工审批环节。
+
+## 价格
+
+| 模型 | 输入（Credits/Token） | 缓存写入（Credits/Token） | 缓存读取（Credits/Token） | 输出（Credits/Token） |
+| :--- | ---: | ---: | ---: | ---: |
+| Muse Spark 1.3 | 1.25 | 1.25 | 0.15 | 4.25 |
diff --git a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/pricing-and-usage.md b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/pricing-and-usage.md
index da05361..c31e61f 100644
--- a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/pricing-and-usage.md
+++ b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/pricing-and-usage.md
@@ -34,6 +34,7 @@
 | Qwen3.8-Max | 2.00 | 2.00 | 0.25 | 6.00 | - |
 | Qwen3.7-Max | 1.65 | 1.65 | 0.33 | 4.951 | - |
 | Qwen3.6-27B | 0.19 | 0.19 | 0.019 | 2.99 | - |
+| Hy4 Preview | 0.834 | 0.834 | 0.042 | 2.501 | - |
 | Hy3 | 0.132 | 0.132 | 0.033 | 0.528 | - |
 | MiMo-V2.5-Pro | 0.435 | 0.435 | 0.0036 | 0.87 | - |
 | MiMo-V2.5 | 0.14 | 0.14 | 0.0028 | 0.28 | - |
@@ -60,6 +61,7 @@
 | GPT-5.4 Nano | 0.20 | 0.20 | 0.02 | 1.25 | 10,000 |
 | GPT-5 Nano | 0.05 | 0.05 | 0.005 | 0.40 | - |
 | Claude Opus 5 | 5.00 | 6.25 | 0.50 | 25.00 | 10,000 |
+| Claude Fable 5.1 | 10.00 | 12.50 | 0.25 | 50.00 | - |
 | Claude Fable 5 | 10.00 | 12.50 | 1.00 | 50.00 | 10,000 |
 | Claude Opus 4.8 | 5.00 | 6.25 | 0.50 | 25.00 | 10,000 |
 | Claude Opus 4.7 | 5.00 | 6.25 | 0.50 | 25.00 | 10,000 |
@@ -69,6 +71,8 @@
 | Claude Sonnet 4.6 | 3.00 | 3.75 | 0.30 | 15.00 | 10,000 |
 | Claude Sonnet 4.5 | 3.00 | 3.75 | 0.30 | 15.00 | 10,000 |
 | Claude Haiku 4.5 | 1.00 | 1.25 | 0.10 | 5.00 | 10,000 |
+| Muse Spark 1.3 | 1.25 | 1.25 | 0.15 | 4.25 | - |
+| Gemini 3.8 Flash | 0.75 | 0.75 | 0.075 | 3.75 | 14,000 |
 | Gemini 3.6 Flash | 1.50 | 1.50 | 0.15 | 7.50 | 14,000 |
 | Gemini 3.5 Flash | 1.50 | 1.50 | 0.15 | 9.00 | 14,000 |
 | Gemini 3.5 Flash-Lite | 0.30 | 0.30 | 0.03 | 2.50 | 14,000 |
@@ -76,7 +80,7 @@
 | Gemini 3 Flash | 0.50 | 0.50 | 0.05 | 3.00 | 14,000 |
 
 :::caution 价格总表说明
-价格总表展示每个模型当前生效的标准参考价。DeepSeek V4 Pro、DeepSeek-V4-Flash 和 DeepSeek-V4-Flash-Vision-Exp 支持分时计费，因此表中同时展示**闲时**与**忙时**价格。根据 DeepSeek 官方最新规则，自 2026 年 8 月 23 日 00:00（UTC+8）起，周六、周日全天按闲时价格计费；工作日具体适用时段及最终账单以平台页面展示为准。DeepSeek-V4-Flash 当前在 B.AI Chat 和 API 中免费，DeepSeek-V4-Flash-Vision-Exp 当前在 B.AI API 中免费；两者在表中均展示标准参考价。MiMo-V2.5 当前 API 调用免费，B.AI Chat 免费开放日期为 2026 年 8 月 25 日，具体开放时间以实际生效为准；其表格行展示标准参考价。表中的“缓存写入（Cache Write）”表示发生缓存写入时的计费价格，不代表所有模型使用统一的缓存有效期。不同模型厂商的缓存策略、缓存有效期、长上下文价格和扩展缓存能力可能不同；如模型存在特殊缓存规则、长上下文价格、1 小时缓存写入价格或分时间生效的价格，请以对应模型详情页说明为准。
+价格总表展示每个模型当前生效的标准参考价。DeepSeek V4 Pro、DeepSeek-V4-Flash 和 DeepSeek-V4-Flash-Vision-Exp 支持分时计费，因此表中同时展示**闲时**与**忙时**价格。DeepSeek API 调用按北京时间（UTC+8）实行分时计费：忙时为周一至周五 09:00-12:00 和 14:00-18:00；其余时间（包括周六、周日全天）为闲时。B.AI Chat 中的 DeepSeek 模型统一按闲时价格计费。自 2026 年 9 月 3 日 17:00（UTC+8）起，DeepSeek-V4-Flash 和 DeepSeek-V4-Flash-Vision-Exp 符合条件的 API 使用按调用发生时对应闲时或忙时标准价的 50% 结算；两者在表中继续展示标准参考价。MiMo-V2.5 当前 API 调用免费，B.AI Chat 免费开放日期为 2026 年 8 月 25 日，具体开放时间以实际生效为准；其表格行展示标准参考价。表中的“缓存写入（Cache Write）”表示发生缓存写入时的计费价格，不代表所有模型使用统一的缓存有效期。不同模型厂商的缓存策略、缓存有效期、长上下文价格和扩展缓存能力可能不同；如模型存在特殊缓存规则、长上下文价格、1 小时缓存写入价格或分时间生效的价格，请以对应模型详情页说明为准。
 :::
 
 :::info 价格说明
diff --git a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/promotions-and-pricing-notices.md b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/promotions-and-pricing-notices.md
index 7961a7e..e130009 100644
--- a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/promotions-and-pricing-notices.md
+++ b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/promotions-and-pricing-notices.md
@@ -12,31 +12,30 @@ import ActivityCard from '@site/src/components/ActivityCard';
 
 <ActivityCard
   variant="free"
-  title="DeepSeek-V4-Flash"
+  title="GLM-5.3-Flash"
   status="免费活动"
-  detail="0 Credits"
+  detail="API 已免费 · Chat 上架后免费"
 >
-活动开始时间：2026 年 8 月 17 日。
-
-本活动覆盖 B.AI Chat 和 API：
+本活动覆盖 B.AI API 和 Chat：
 
-* **Chat：** 活动期间使用 DeepSeek-V4-Flash，按 `0 Credits` 结算。
-* **API：** 活动期间使用按 `0 Credits` 结算，不收取单次请求、输入、缓存写入、缓存读取或输出的费用。
+* **API：** GLM-5.3-Flash API 调用目前按 `0 Credits` 结算，不收取输入、缓存写入、缓存读取或输出 Token 费用。
+* **Chat：** GLM-5.3-Flash 在 B.AI Chat 上架后免费开放，具体开放时间以模型实际上架为准；开放后，Chat 使用按 `0 Credits` 结算。
 
-活动结束后，DeepSeek-V4-Flash 将恢复标准价格。详见[模型详情](./models/deepseek-v4-flash.md)。
+活动结束后，GLM-5.3-Flash 将恢复[模型详情](./models/glm-5-3-flash.md)中展示的价格。
 </ActivityCard>
 
 <ActivityCard
   variant="free"
-  title="DeepSeek-V4-Flash-Vision-Exp"
-  status="API 免费"
-  detail="0 Credits"
+  title="Qwen3.8-Flash"
+  status="免费活动"
+  detail="API 已免费 · Chat 上架后免费"
 >
-本活动当前生效，仅适用于通过 B.AI API 发起的 `deepseek-v4-flash-vision-exp` 调用。
+B.AI API 和 Chat 分阶段开放免费使用：
 
-活动期间，API 调用按 `0 Credits` 结算，不收取单次请求、输入、缓存写入、缓存读取或输出费用。
+* **API：** Qwen3.8-Flash API 调用目前按 `0 Credits` 结算，不收取输入、缓存写入、缓存读取或输出 Token 费用。
+* **Chat：** Qwen3.8-Flash 在 B.AI Chat 上架后免费开放，具体免费开放日期以模型实际上架为准；开放后，Chat 使用按 `0 Credits` 结算。
 
-活动结束后，该模型将恢复标准价格；标准参考价请查看[定价与用量](./pricing-and-usage.md)。
+活动结束后，Qwen3.8-Flash 将恢复[模型详情](./models/qwen3-8-flash.md)中展示的价格。
 </ActivityCard>
 
 <ActivityCard
@@ -70,31 +69,29 @@ MiMo-V2.5 免费分阶段开放：
 </ActivityCard>
 
 <ActivityCard
-  variant="free"
-  title="GLM-5.3-Flash"
-  status="免费活动"
-  detail="API 已免费 · Chat 上架后免费"
+  variant="discount"
+  title="DeepSeek-V4-Flash"
+  status="API 限时折扣"
+  detail="标准价 5 折"
 >
-本活动覆盖 B.AI API 和 Chat：
+DeepSeek-V4-Flash 5 折活动于 2026 年 9 月 3 日 17:00（UTC+8）正式生效。
 
-* **API：** GLM-5.3-Flash API 调用目前按 `0 Credits` 结算，不收取输入、缓存写入、缓存读取或输出 Token 费用。
-* **Chat：** GLM-5.3-Flash 在 B.AI Chat 上架后免费开放，具体开放时间以模型实际上架为准；开放后，Chat 使用按 `0 Credits` 结算。
+自生效时间起，符合条件的 DeepSeek-V4-Flash API 使用按对应闲时或忙时标准价的 50% 结算。活动价格随 DeepSeek 闲时与忙时计费时段同步变化，各时段均保持 5 折。
 
-活动结束后，GLM-5.3-Flash 将恢复[模型详情](./models/glm-5-3-flash.md)中展示的价格。
+[定价与用量](./pricing-and-usage.md)价格总表和[模型详情](./models/deepseek-v4-flash.md)继续展示标准参考价，实际结算价格及最终账单以平台页面展示为准。
 </ActivityCard>
 
 <ActivityCard
-  variant="free"
-  title="Qwen3.8-Flash"
-  status="免费活动"
-  detail="API 已免费 · Chat 上架后免费"
+  variant="discount"
+  title="DeepSeek-V4-Flash-Vision-Exp"
+  status="API 限时折扣"
+  detail="标准价 5 折"
 >
-B.AI API 和 Chat 分阶段开放免费使用：
+DeepSeek-V4-Flash-Vision-Exp 5 折活动于 2026 年 9 月 3 日 17:00（UTC+8）正式生效。
 
-* **API：** Qwen3.8-Flash API 调用目前按 `0 Credits` 结算，不收取输入、缓存写入、缓存读取或输出 Token 费用。
-* **Chat：** Qwen3.8-Flash 在 B.AI Chat 上架后免费开放，具体免费开放日期以模型实际上架为准；开放后，Chat 使用按 `0 Credits` 结算。
+自生效时间起，符合条件的 `deepseek-v4-flash-vision-exp` API 使用按对应闲时或忙时标准价的 50% 结算。活动价格随 DeepSeek 闲时与忙时计费时段同步变化，各时段均保持 5 折。
 
-活动结束后，Qwen3.8-Flash 将恢复[模型详情](./models/qwen3-8-flash.md)中展示的价格。
+[定价与用量](./pricing-and-usage.md)价格总表和[模型详情](./models/deepseek-v4-flash-vision-exp.md)继续展示标准参考价，实际结算价格及最终账单以平台页面展示为准。
 </ActivityCard>
 
 <ActivityCard
@@ -131,7 +128,7 @@ B.AI API 和 Chat 分阶段开放免费使用：
   status="分时计费"
   detail="闲时 / 忙时价格"
 >
-DeepSeek-V4-Pro 支持闲时与忙时分时计费，具体适用时段及最终账单以平台页面展示为准。
+DeepSeek API 调用按北京时间（UTC+8）实行分时计费：忙时为周一至周五 09:00-12:00 和 14:00-18:00；其余时间（包括周六、周日全天）为闲时。B.AI Chat 中的 DeepSeek 模型统一按闲时价格计费。
 
-DeepSeek-V4-Flash 同样支持闲时与忙时分时计费，但当前处于限时免费活动，B.AI Chat 和 API 使用均免费。标准价格请查看 [DeepSeek-V4-Pro](./models/deepseek-v4-pro.md) 和 [DeepSeek-V4-Flash](./models/deepseek-v4-flash.md) 模型详情。最终账单以平台页面展示为准。
+DeepSeek-V4-Pro、DeepSeek-V4-Flash 和 DeepSeek-V4-Flash-Vision-Exp 均适用上述闲时与忙时规则。自 2026 年 9 月 3 日 17:00（UTC+8）起，符合条件的 DeepSeek-V4-Flash 和 DeepSeek-V4-Flash-Vision-Exp API 使用按调用发生时对应闲时或忙时标准价的 50% 结算。标准价格请查看 [DeepSeek-V4-Pro](./models/deepseek-v4-pro.md) 和 [DeepSeek-V4-Flash](./models/deepseek-v4-flash.md) 模型详情。最终账单以平台页面展示为准。
 </ActivityCard>
diff --git a/package.json b/package.json
index 9d06639..fddafe4 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@x402-tron/docs",
-  "version": "1.3.26",
+  "version": "1.3.27",
   "description": "x402-tron documentation",
   "license": "MIT",
   "resolutions": {
diff --git a/sidebars.js b/sidebars.js
index ed536d5..57d330d 100644
--- a/sidebars.js
+++ b/sidebars.js
@@ -173,6 +173,7 @@ const sidebars = {
               items: [
                 { type: 'doc', id: 'llmservice/models/claude-opus-5', label: 'Claude Opus 5' },
                 { type: 'doc', id: 'llmservice/models/claude-sonnet-5', label: 'Claude Sonnet 5' },
+                { type: 'doc', id: 'llmservice/models/claude-fable-5-1', label: 'Claude Fable 5.1' },
                 { type: 'doc', id: 'llmservice/models/claude-fable-5', label: 'Claude Fable 5' },
                 { type: 'doc', id: 'llmservice/models/claude-opus-4-8', label: 'Claude Opus 4.8' },
                 { type: 'doc', id: 'llmservice/models/claude-opus-4-7', label: 'Claude Opus 4.7' },
@@ -188,6 +189,7 @@ const sidebars = {
               label: 'Google (Gemini)',
               collapsed: true,
               items: [
+                { type: 'doc', id: 'llmservice/models/gemini-3-8-flash', label: 'Gemini 3.8 Flash' },
                 { type: 'doc', id: 'llmservice/models/gemini-3-6-flash', label: 'Gemini 3.6 Flash' },
                 { type: 'doc', id: 'llmservice/models/gemini-3-5-flash', label: 'Gemini 3.5 Flash' },
                 { type: 'doc', id: 'llmservice/models/gemini-3-5-flash-lite', label: 'Gemini 3.5 Flash-Lite' },
@@ -195,6 +197,14 @@ const sidebars = {
                 { type: 'doc', id: 'llmservice/models/gemini-3-flash', label: 'Gemini 3 Flash' },
               ],
             },
+            {
+              type: 'category',
+              label: 'Meta (Muse)',
+              collapsed: true,
+              items: [
+                { type: 'doc', id: 'llmservice/models/muse-spark-1-3', label: 'Muse Spark 1.3' },
+              ],
+            },
             {
               type: 'category',
               label: 'DeepSeek',
@@ -202,6 +212,7 @@ const sidebars = {
               items: [
                 { type: 'doc', id: 'llmservice/models/deepseek-v4-pro', label: 'DeepSeek V4 Pro' },
                 { type: 'doc', id: 'llmservice/models/deepseek-v4-flash', label: 'DeepSeek-V4-Flash' },
+                { type: 'doc', id: 'llmservice/models/deepseek-v4-flash-vision-exp', label: 'DeepSeek-V4-Flash-Vision-Exp' },
                 { type: 'doc', id: 'llmservice/models/deepseek-v3.2', label: 'DeepSeek V3.2' },
               ],
             },
@@ -210,6 +221,7 @@ const sidebars = {
               label: 'Tencent Hunyuan (Hy)',
               collapsed: true,
               items: [
+                { type: 'doc', id: 'llmservice/models/hy4-preview', label: 'Hy4 Preview' },
                 { type: 'doc', id: 'llmservice/models/hy3', label: 'Hy3' },
               ],
             },