[BUG] OpenCode Go returns HTTP 500 for GPT-5.6 Luna via /chat/completions; model works via /responses

[AntzCode]

Problem (one or two sentences)

Zoo Code in code mode using gpt-5.6-luna with a valid Opencode Go api key has recently started responding with 500 Provider Error. This started occurring approximately 20260827T000000. It was previously working without any problems.

zoo-diagnostics-01a0498b-1787940297118.json

Context (who is affected and when)

When using Opencode Go subscription with gpt-5.6-luna model, a 500 Provider Error occurs. Same key and model succeeds when using the opencode app on the same projects on the same host.

Further testing has identified the same problem for basic curl requests to the api endpoint /chat/completions, whereas the same parameters will produce a working response on the /responses endpoint.

I have attached files with minimal curl examples. NB: Opencode Go api key is required for these tests, please contact me privately if you require a temporary key for testing.

example-curl-ok.txt
example-curl-error.txt

Reproduction steps

Using VS Codium 1.26.04524 and Zoo Code 3.68.0 (1833f5a).

• Click on Zoo Code > Models, add a model.
• Select provider "Opencode Go"
• Enter your Opencode Go API key
• Click refresh models button
• Select gpt-5.6-luna
• Save settings
• Prompt Zoo Code using the gpt-5.6-luna model

Expected result

Zoo Code should show thinking and report its actions

Actual result

It shows a 500 Provider error

Variations tried (optional)

Tried with two separate Opencode Go subscriptions:
Subscription 1 for work on work machine Ubuntu 26.04 (Mate) + VS Codium + Zoo Code.
Subscription 2 for home on home machine Linux Mint 22.2 (Mate) + VS Codium + Zoo Code.
Both give the 500 Provider error.
Tried with opencode app using work api key at work, succeeds.
Tried with opencode app using home api key at home, succeeds.
Tried basic curl request on /chat/completions endpoint using home key, fails.
Tried basic curl request on /responses endpoint using home key, succeeds.

After LLM analysis of Zoo Code Github repository compared to opencode Github repository, LLM advises that opencode appears to be using the /responses endpoint whereas Zoo Code is using the /chat/completions endpoint, see screenshot:

App Version

Version: 3.68.0 (1833f5a)

API Provider (optional)

OpenAI Compatible

Model Used (optional)

gpt-5.6-luna

Zoo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[AntzCode]

Claiming. I already have a working fix on a local checked-out out fork. I will follow contribution guidelines to commit and submit a pull request for this.