Problem

On the opencode-go provider, qwen3.7-max and minimax-m2.7 fail to produce any response. Both models appear in the /models catalog (16 models total), but actual API calls fail.

Reproduction

qwen3.7-max

hermes chat -q "Say OK" --provider opencode-go -m qwen3.7-max -Q
# HTTP 404 — Not Found | opencode

# Direct OpenAI SDK:
# Error code: 401 — Model qwen3.7-max is not supported for format oa-compat

minimax-m2.7

hermes chat -q "Say OK" --provider opencode-go -m minimax-m2.7 -Q
# HTTP 404 — Not Found | opencode

# Direct OpenAI SDK:
# Error: 'NoneType' object is not subscriptable (malformed response)

# Direct Anthropic SDK (same base_url):
# Returns 404 HTML page (proxy does not route /v1/messages for this model)

Working Models (for reference)

These models work fine on the same opencode-go provider:

• kimi-k2.6 (with fix from #35180)
• deepseek-v4-pro
• glm-5.1
• qwen3.6-plus

Analysis

The OpenCode Go proxy (opencode.ai/zen/go/v1) only exposes the OpenAI-compatible chat completions format. qwen3.7-max and minimax-m2.7 apparently require a different wire format (possibly Anthropic Messages) but the Go proxy does not expose an Anthropic-compatible endpoint for them.

Notes

• The OpenCodeGoProfile transport is configured as openai_chat — this is correct for kimi/deepseek/glm/qwen3.6.
• minimax-m2.7 is documented in the OpenCode Go source comments as using anthropic_messages transport, but no Anthropic endpoint exists at /zen/go/v1.
• This is separate from the kimi thinking/reasoning conflict (#32223 / #35180).

Environment

• Hermes Agent v2026.5.29 (current main)
• Provider: opencode-go (OpenCode Go subscription)
• Affected models: qwen3.7-max, minimax-m2.7