Models
One key reaches every model — Claude (Fable, Opus, Sonnet, Haiku), GPT and Codex, GLM, DeepSeek, Kimi, MiniMax, Qwen, NVIDIA Nemotron, and more. Select a model with the usual model field your client already sends; switch models without changing keys or accounts. You pay native per-token rates with 0% markup (the Auto routing ids are the one flat-priced exception) — the live catalog and rates are on the Models section and the portal.
List what your key reaches
This snapshot drifts; the API never does. The .data[].id shape works on both the Anthropic and OpenAI /v1/models responses:
curl -s https://app.inferencehub.tech/v1/models \ -H "Authorization: Bearer sk-prov-live-YOUR_KEY" | jq -r '.data[].id'Model ids & aliases
- Short aliases — Claude models have stable short ids that track the newest version:
sonnet,opus,haiku, plus versioned ones likesonnet-5,opus-5,fable-5. Canonical ids (claude-sonnet-4-6) work too. - Vendor-prefixed aliases — open-weight models also answer to vendor-prefixed ids (
zai-glm-5.2,moonshotai-kimi-k2.6). Same model, same price — they exist because some clients (Cursor) hijack the bare names to their own providers. - Claude Code aliases — with gateway model discovery on, the open-weight daily drivers appear in Claude Code's
/modelpicker asclaude-code-glm-5.2,claude-code-deepseek-3.2, … — again the same models at the same prices.
Which agent reaches which models
Each agent speaks one wire format, and the gateway's translation lanes bridge the gaps:
| Agent | Wire | Reaches |
|---|---|---|
| Claude Code | Anthropic Messages | Claude (Opus, Sonnet, Haiku, …) + open-weight daily drivers (GLM, DeepSeek, Kimi, MiniMax, Qwen) |
| Kilo Code | Anthropic Messages + OpenAI Chat | Full catalog — Claude + open-weight on the Anthropic provider; chat catalog + GPT-5.6 on the OpenAI-wire provider |
| Codex | OpenAI Responses | GPT & Codex models (gpt-5.x-codex, GPT-5.6, GPT-5.4 family) |
| Cursor | OpenAI Chat | Open-weight (gpt-oss, GLM, DeepSeek, Kimi, Nemotron) + Claude (sonnet, opus, …) + GPT-5.6 via the gateway's translation lanes |
| ZCode | Anthropic Messages | Claude, plus open-weight daily drivers — GLM-5.2 is its home model |
| Hermes | OpenAI Chat | Open-weight only — gpt-oss, DeepSeek, Kimi, Nemotron, GLM |
Common ids at a glance
A snapshot of the most-used ids (the portal always reflects what's enabled right now):
| id | Anthropic wire |
|---|---|
| sonnet | Claude Sonnet 4.6 |
| sonnet-5 | Claude Sonnet 5 |
| opus | Claude Opus 4.8 |
| opus-5 | Claude Opus 5 |
| haiku | Claude Haiku 4.5 |
| fable-5 | Claude Fable 5 |
| id | Open-weight |
|---|---|
| glm-5.2 | GLM-5.2 (Z.ai) |
| deepseek-3.2 | DeepSeek 3.2 |
| kimi-k3 | Kimi K3 (Moonshot AI) |
| kimi-k2.6 | Kimi K2.6 (Moonshot AI) |
| minimax-m2.5 | MiniMax M2.5 |
| qwen3.5-397b-a17b | Qwen3.5 397B-A17B |
GPT ids: gpt-5.6-sol / gpt-5.6-terra / gpt-5.6-luna and the gpt-5.x-codex family serve on the Responses wire (Codex; GPT-5.6 also reaches Cursor); gpt-oss-120b and the GPT-5.4 chat models serve on the Chat wire.
Premium vs standard models
On subscription plans, the Claude and GPT models priced $4+ per 1M input tokens (Opus, GPT-5.6 Sol, Fable 5, GPT-5.5) draw from one shared monthly premium budget; every other model — open-weight and lower-priced frontier alike (Sonnet, GPT-5.4, GPT-5.6 Terra/Luna) — draws from the larger standard windows. Details on Billing & plans.