Models

Consortium runs models on two independent planes. Which one you’re on decides both the model list and the bill.

Plane	Model list	Who bills you
Managed	The catalog below	Consortium, from your credit wallet
BYOK	Whatever your own account or subscription offers	Your provider, directly

The live managed catalog is served at GET /v1/models — no auth required. The app filters its pickers against it, so an entry below can be temporarily hidden when an upstream provider is unavailable.

Managed catalog

Prices are per 1M tokens, in US dollars. These are provider list prices; managed billing applies a markup on top (see Pricing & credits).

Anthropic

Model	Context	Input	Cached input	Output
claude-opus-4-7	200K	$15.00	$1.50	$75.00
claude-sonnet-4-6	200K	$3.00	$0.30	$15.00
claude-haiku-4-5-20251001	200K	$1.00	$0.10	$5.00

OpenAI

Model	Context	Input	Cached input	Output
gpt-5	—	$1.25	$0.13	$10.00
gpt-5-codex	—	$1.25	—	$10.00
gpt-5-mini	—	$0.25	$0.03	$2.00
gpt-5-nano	—	$0.05	$0.01	$0.40
o4-mini	—	$1.10	$0.28	$4.40

Google

Model	Context	Input	Cached input	Output
gemini-2.5-pro	1M	$1.25	—	$10.00
gemini-2.5-flash	1M	$0.15	—	$0.60

A dash means the value isn’t published for that model. Cached input applies only where the provider supports prompt caching; Consortium bills cache reads and cache writes separately, and input token counts always exclude cached tokens.

Note

Context windows are listed only where they’re confirmed against the provider’s datasheet. An empty cell means Consortium doesn’t pin a number, not that the window is small.

Free models

Available on every plan within the free allowance — 250K tokens per rolling 4 hours on Free. See Pricing & credits.

Model	Notes
big-pickle	Default cold-start model for Consortium Code
nemotron-3-super-free
cloud-free	Rotating free model; falls back automatically

Free-model usage is metered but never debits your wallet.

Caution

cloud-free points at whichever free upstream is currently available. When a provider delists its free tier, the alias re-points — sessions pinned to it can change model between runs.

Model IDs and aliases

Some IDs track a moving upstream instead of a fixed revision.

• claude-opus-4-7 serves the latest Opus 4.x upstream and is not pinned to a revision. Use it when you want automatic minor upgrades.
• Fully dated IDs like claude-haiku-4-5-20251001 pin one release. Use these when a workflow needs byte-stable behavior across runs.
• cloud-free is a rotating alias, not a model — see the caution above.

Pin a dated ID for anything you benchmark, evaluate, or bill a customer for. Track a moving ID for everything else.

Per-agent model pickers

Agents that wrap a vendor CLI expose that vendor’s model list, not the managed catalog. These run under BYOK or the agent’s own subscription auth, so they’re billed by the provider and priced by your plan with them — not by the table above.

Agent	Models offered	Default
Claude Code	default, claude-opus-4-8, claude-opus-4-7, claude-sonnet-5, claude-sonnet-4-5, claude-haiku-4-5, claude-fable-5, adaptive	default (CLI decides)
Codex	gpt-5.5, gpt-5.4, gpt-5.3-codex (each at minimal/low/medium/high effort), gpt-5.4-mini, gpt-5.3-codex-spark, gpt-5.2	gpt-5.5 (medium)
Gemini	gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-lite	gemini-2.5-pro
Grok	grok-build, grok-code-fast-1, grok-4, grok-4-fast	grok-build
Consortium Code	Free models above + the managed catalog	big-pickle
Pi	Anthropic models	claude-opus-4-8

Note

The two lists differ on purpose. A model in a CLI picker but not in the managed catalog — claude-sonnet-5, for instance — runs fine under your own subscription; it just can’t be paid for with credits.

Pickers filter live: options that need credits are labeled, and options from your connected accounts appear once the machine’s CLI has probed them.

Picking a model

Use the agent’s default unless you have a reason not to. For long, tool-heavy runs the small variants — haiku, a mini or spark class, flash — usually hold up well at a fraction of the cost.