[BUG] deepseek-v4-flash via opencode-go returns 403 / hangs, while deepseek-v4-pro and minimax-m3 work on the same key

Environment

• Windows 11, OpenCode Desktop + TUI (latest)
• Provider: opencode-go (Go plan)

Config status

• Global config (~/.config/opencode/opencode.jsonc): clean, only $schema
• No project-level opencode.json overrides
• opencode-go API key present and valid

Repro steps

1. Send a chat request with model deepseek-v4-flash to https://opencode.ai/zen/go/v1/chat/completions
2. Observed: request hangs with no response (tested up to 60s+), and/or returns HTTP 403

Direct API test results (same key, same endpoint, same request body)

Log evidence

(~/.local/share/opencode/log/opencode.log)

• stream error ... providerID=opencode modelID=laguna-s-2.1-free ... error.error="AI_APICallError: Internal Server Error" (free route also affected earlier)
• No client-side config or auth errors; network to opencode.ai:443 (Cloudflare) is fine

Impact

• Requests to deepseek-v4-flash never complete in the UI (waited ~11 minutes), while free models and other Go paid models respond normally
• small_model / title requests also fail when set to flash

Related: #40465, #40479, #40480

Begin with the direct requests to https://opencode.ai/zen/go/v1/chat/completions, comparing deepseek-v4-flash with deepseek-v4-pro and minimax-m3 using the same key and body. Review ~/.local/share/opencode/log/opencode.log alongside the related issues #40465, #40479, and #40480. Done means establishing whether the failure is in OpenCode's handling or the provider route and documenting a reproducible resolution.