Provider Support and Certification Matrix
Generated on 2026-08-13 from the static runtime catalog and the human-maintained evidence ledger. Do not edit the table manually.
CoreMind currently supports configuration for 40 built-in providers. Complete real-call evidence exists for 1, while 39 are catalog-only. Configurable does not mean certified.
Status definitions
- Certified: the same model passed real streaming, tool-call, structured-result, multi-turn, abort, error-mapping, and long-context checks with auditable evidence.
- Configurable, unverified: configuration and model catalog resolution work, but complete real-call evidence is not available.
- Configurable, incomplete current certification: earlier evidence is retained but does not cover every check in the current seven-check standard.
- Custom OpenAI-compatible endpoints are deployment-specific and must be accepted by each project.
Current matrix
| Provider ID | Default model | Certified version | Models | Status | Evidence |
|---|---|---|---|---|---|
alibaba-model-studio | qwen-plus | 0.3.0 | 1 | Certified | Evidence |
amazon-bedrock | amazon.nova-2-lite-v1:0 | — | 114 | Configurable, unverified | — |
ant-ling | Ling-2.6-1T | — | 3 | Configurable, unverified | — |
anthropic | claude-fable-5 | — | 13 | Configurable, unverified | — |
azure-openai-responses | gpt-4 | — | 38 | Configurable, unverified | — |
baseten | deepseek-ai/DeepSeek-V4-Flash-0731 | — | 16 | Configurable, unverified | — |
cerebras | gemma-4-31b | — | 3 | Configurable, unverified | — |
cloudflare-ai-gateway | claude-3-5-haiku | — | 43 | Configurable, unverified | — |
cloudflare-workers-ai | @cf/google/gemma-4-26b-a4b-it | — | 13 | Configurable, unverified | — |
deepseek | deepseek-v4-flash | — | 2 | Configurable, unverified | — |
fireworks | accounts/fireworks/models/deepseek-v4-flash | — | 17 | Configurable, unverified | — |
github-copilot | claude-haiku-4.5 | — | 30 | Configurable, unverified | — |
google | deep-research-max-preview-04-2026 | — | 24 | Configurable, unverified | — |
google-vertex | gemini-2.5-flash | — | 12 | Configurable, unverified | — |
groq | llama-3.1-8b-instant | — | 6 | Configurable, unverified | — |
huggingface | MiniMaxAI/MiniMax-M2 | — | 58 | Configurable, unverified | — |
kimi-coding | k3 | — | 4 | Configurable, unverified | — |
minimax | MiniMax-M2.7 | — | 3 | Configurable, unverified | — |
minimax-cn | MiniMax-M2.7 | — | 3 | Configurable, unverified | — |
mistral | codestral-latest | — | 31 | Configurable, unverified | — |
moonshotai | kimi-k2-0711-preview | — | 10 | Configurable, unverified | — |
moonshotai-cn | kimi-k2-0711-preview | — | 10 | Configurable, unverified | — |
nvidia | google/gemma-3-12b-it | — | 30 | Configurable, unverified | — |
openai | gpt-4 | — | 38 | Configurable, unverified | — |
openai-codex | gpt-5.3-codex-spark | — | 7 | Configurable, unverified | — |
opencode | claude-fable-5 | — | 59 | Configurable, unverified | — |
opencode-go | minimax-m3 | — | 18 | Configurable, unverified | — |
openrouter | ai21/jamba-large-1.7 | — | 335 | Configurable, unverified | — |
qwen-token-plan | MiniMax-M2.5 | — | 16 | Configurable, unverified | — |
qwen-token-plan-cn | MiniMax-M2.5 | — | 16 | Configurable, unverified | — |
qwen-token-plan-individual | deepseek-v4-flash-0731 | — | 7 | Configurable, unverified | — |
together | MiniMaxAI/MiniMax-M2.7 | — | 18 | Configurable, unverified | — |
vercel-ai-gateway | alibaba/qwen-3-14b | — | 197 | Configurable, unverified | — |
xai | grok-4.3 | — | 3 | Configurable, unverified | — |
xiaomi | mimo-v2-flash | — | 6 | Configurable, unverified | — |
xiaomi-token-plan-ams | mimo-v2-pro | — | 3 | Configurable, unverified | — |
xiaomi-token-plan-cn | mimo-v2-pro | — | 3 | Configurable, unverified | — |
xiaomi-token-plan-sgp | mimo-v2-pro | — | 3 | Configurable, unverified | — |
zai | glm-4.7 | — | 4 | Configurable, unverified | — |
zai-coding-cn | glm-4.7 | — | 4 | Configurable, unverified | — |
Adding a certification
Follow the Provider certification SOP. Add records only to certifications.json, then run npm run providers:matrix. Never mark a provider as certified without real credentials, run logs, and error-path evidence.