Skip to content
FoundationMachines

Models

Frontier models served by Foundation Machines, and how Sebastion AI picks one for you.

Models

Foundation Machines exposes a curated set of frontier models. The /v1/chat/completions gateway lets you pick a model with the model parameter; the /v1/audit endpoint and the Sebastion AI GitHub App choose one for you based on your plan.

Frontier models

VendorModelContextAPI idStatus
AnthropicClaude Haiku 4.5200Kclaude-haiku-4.5Available
AnthropicClaude Sonnet 4.6200Kclaude-sonnet-4.6Available
AnthropicClaude Opus 4.7200Kclaude-opus-4.7Available
OpenAIGPT-5.4256Kgpt-5.4Available
OpenAIGPT-5.4 mini256Kgpt-5.4-miniAvailable
GoogleGemini 3 Flash1Mgemini-3-flashAvailable
MoonshotKimi K2.5200Kkimi-k2.5Forthcoming

For /v1/chat/completions (the gateway)

When calling the chat-completions gateway you pass model explicitly. The set you can choose from depends on your plan tier.

PlanAllowed models
Freeclaude-haiku-4.5, gpt-5.4-mini, gemini-3-flash
Proadds claude-sonnet-4.6, gpt-5.4
Teamadds claude-opus-4.7
Enterprisefull set, including any forthcoming models on launch

Calls with a model outside your plan return 403 model_forbidden.

For /v1/audit and the Sebastion AI GitHub App

Sebastion AI does not take a model parameter. You upgrade your plan; Sebastion routes you to a stronger model automatically.

PlanAudit model
Freeclaude-sonnet-4.6
Proclaude-opus-4.7
Teamclaude-opus-4.7 (multi-model cross-check coming)
Enterprisebest frontier model + on-prem option

This applies to both the GitHub App and direct POST /v1/audit calls.

The live catalogue

Beyond the headline models above we may serve additional variants for testing. List the live catalogue:

curl https://api.foundationmachines.ai/v1/models \
  -H "Authorization: Bearer fm_..."

The data array returned by /v1/models is the source of truth for what is currently routable. Plans gate which subset of that catalogue you can actually call from /v1/chat/completions.