[Bug]: OpenCode Zen routes kimi-k2.5-free to /messages instead of /chat/completions

Bug Description

When using OpenCode Zen as a provider with kimi-k2.5-free, Hermes routes the request to https://opencode.ai/zen/v1/messages (the Anthropic/Claude endpoint) instead of https://opencode.ai/zen/v1/chat/completions (the correct endpoint for Kimi models).

This results in a 404 Not Found error every time, regardless of config changes.

According to OpenCode Zen documentation, different model families require different endpoints:

The opencode-zen provider appears to default all models to the /messages endpoint rather than routing based on model family.

Steps to Reproduce

1. Configure OpenCode Zen in ~/.hermes/.env:

   OPENCODE_ZEN_API_KEY=sk-...

2. Set model in config.yaml:

   model: default: kimi-k2.5-free provider: opencode-zen

   No base_url or api_mode override — clean config, provider-only.

3. Run hermes and type any message (e.g., hello).

4. Observe the error:

   ⚠️ API call failed (attempt 1/3): NotFoundError [HTTP 404] 🔌 Provider: opencode-zen Model: kimi-k2.5-free 🌐 Endpoint: https://opencode.ai/zen/v1/messages 📝 Error: HTTP 404 — Not Found | opencode

Expected Behavior

The opencode-zen provider should detect that kimi-k2.5-free is a Kimi model and route the request to https://opencode.ai/zen/v1/chat/completions instead of https://opencode.ai/zen/v1/messages.

Workaround

Manually forcing the endpoint in config.yaml:

model: default: kimi-k2.5-free provider: opencode-zen base_url: https://opencode.ai/zen/v1/chat/completions api_mode: chat_completions

This works but defeats the purpose of automatic provider routing.

Environment

• OS: Windows 11 / WSL2 (Ubuntu)
• Hermes version: v0.10.0 (also reproduced after hermes update to upstream ac4e8cb4)
• Python version: 3.11
• Model: kimi-k2.5-free
• Provider: opencode-zen

Related Issues

• #7710 — Similar routing issue with minimax-m2.5-free via OpenCode Zen

Additional Context

This likely affects all non-Claude, non-Anthropic models served through OpenCode Zen (Kimi, GLM, Qwen, MiniMax, Big Pickle, Gemini, GPT). The provider seems to unconditionally use the Anthropic /messages endpoint format rather than selecting the correct API format based on the model ID.

TL;DR

The opencode-zen provider defaults to the /messages endpoint instead of routing based on the model family, causing a 404 error when using Kimi models.

Guidance

• Verify that the model family is correctly identified as Kimi for the kimi-k2.5-free model.
• Check the opencode-zen provider configuration to ensure it is not hardcoded to use the /messages endpoint.
• Consider adding a workaround by manually forcing the endpoint in config.yaml as shown in the issue description.
• Review the OpenCode Zen documentation to ensure that the correct endpoint is being used for Kimi models.

Example

model:
  default: kimi-k2.5-free
  provider: opencode-zen
  base_url: https://opencode.ai/zen/v1/chat/completions
  api_mode: chat_completions

Notes

The issue seems to affect all non-Claude, non-Anthropic models served through OpenCode Zen. The provider's default behavior may need to be updated to correctly route requests based on the model family.

Recommendation

Apply the workaround by manually forcing the endpoint in config.yaml until a permanent fix is available, as it allows the model to function correctly despite the provider's default behavior.