This document records the minimum validation gate for curated GonkaGate models and the approved public validated models in the shipped runtime.
A model may be marked validated only after end-to-end verification against
the current verified OpenCode baseline:
- interactive
opencode opencode run- streaming text responses
- tool-calling and file-edit loops
- multi-turn continuation
- the
small_modelpath used by lightweight OpenCode tasks - effective-config resolution in both
userandprojectscope - any required provider options, model options, or model headers needed for stable validated behavior
The current shipped runtime exposes these validated models through the public picker.
- display name:
Qwen3 235B A22B Instruct 2507 FP8 - key:
qwen3-235b-a22b-instruct-2507-fp8 - model id:
qwen/qwen3-235b-a22b-instruct-2507-fp8 - transport:
chat_completions - adapter package:
@ai-sdk/openai-compatible - recommended:
false
- display name:
Kimi K2.6 - key:
kimi-k2.6 - model id:
moonshotai/Kimi-K2.6 - transport:
chat_completions - adapter package:
@ai-sdk/openai-compatible - recommended:
true
- display name:
MiniMax M2.7 - key:
minimax-m2.7 - model id:
minimaxai/minimax-m2.7 - transport:
chat_completions - adapter package:
@ai-sdk/openai-compatible - recommended:
false
- v1 explicitly writes both
modelandsmall_modelto the same selected validated GonkaGate model - v1 writes every validated curated model into
provider.gonkagate.modelsso OpenCode's/modelscommand can switch between managed GonkaGate models - interactive setup keeps the public curated picker visible as broader public model choice lands without changing the UX contract
- future
/v1/responsessupport requires a separate migration and revalidation - broader public model choice stays gated on entries passing the same validation gate