Model catalogue

Models available in v1, by provider. The prices below are the figures the platform uses for its own cost accounting — your provider's bill is authoritative.

Anthropic Claude

Per-token billing. Genesis prices Anthropic models per family and discovers the models themselves dynamically from the Anthropic API: any claude-* model the API serves is accepted and billed at its family's rates. When a new version appears — claude-opus-4-8, say — it works without a platform update and inherits the Opus family's pricing and context window.

FamilyInput ($/MTok)Output ($/MTok)ContextBest for
Fable$10$501M tokensThe hardest work — a tier above Opus.
Opus$5$25200KComplex reasoning, architecture, strategy.
Sonnet$3$15200KBalanced. Coding, analysis, general tasks.
Haiku$1$5200KFastest and cheapest. Simple tasks, high volume.

Model IDs look like claude-fable-5, claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5-20251001.

Unless you change it, conversations run on Claude Opus 4.7 (claude-opus-4-7).

Synthetic.new

Flat-rate via subscription — no per-token prices. Model IDs carry the hf: prefix. The registered set:

ModelID
MiniMax M2.5hf:MiniMaxAI/MiniMax-M2.5
Qwen 3.6 27Bhf:Qwen/Qwen3.6-27B
Kimi K2.6hf:moonshotai/Kimi-K2.6
NVIDIA Nemotron 3 Super 120Bhf:nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4
GLM 4.7hf:zai-org/GLM-4.7
GLM 4.7 Flashhf:zai-org/GLM-4.7-Flash
GLM 5.1hf:zai-org/GLM-5.1
GPT OSS 120Bhf:openai/gpt-oss-120b
Qwen3 Coder 480Bhf:Qwen/Qwen3-Coder-480B-A35B-Instruct
Qwen 3.5 397Bhf:Qwen/Qwen3.5-397B-A17B

The full catalogue is fetched live from the Synthetic API, so the model picker may show more than this list.

Other providers

If your deployment has GitHub Copilot or JetBrains Junie connected (bring your own account), their models appear in the model picker as their own provider groups. They are metered by those subscriptions — no per-token prices here either.

Picking, in 30 seconds

  • No idea what to use → leave the defaults.
  • The hardest problems, or very large context → Fable (1M-token context; be ready for the bill).
  • Cost-conscious → Synthetic flat rate; Qwen3 Coder 480B is the coding specialist.
  • High volume, mechanical work → Haiku.

Defaults

If you don't set anything, conversations use claude-opus-4-7. Model presets for the platform's other roles (decomposition, review, and the rest) are seeded by the platform; the LLM Configuration page shows — and lets you edit — what your tenant actually uses. That page is authoritative, not this one.