Models

Every listed model has a fixed price per token, and these are the prices you pay. Free models cost nothing: see Free models. The list below is live from GET /models.

Model idsCopy Anchor Link

Send the id as model, like anthropic/claude-sonnet-5.5. A few more forms work:

• Claude clients' own names on /messages: claude-sonnet-5-5, dated names like claude-haiku-4-5-20251001 and the [1m] suffix map to the listed model.
• Latest aliases like ~anthropic/claude-opus-latest work and are priced as the model they point to.
• Variants change how a model runs, not what it costs per token: :online, :nitro, :floor and :exacto. See Web search and Routing and fallbacks.

Free modelsCopy Anchor Link

Free models cost nothing to call, even with no credit. Send one as model:

• ZeroRoute (binference/free) matches each call to a free model that can do what it needs, such as read images, call tools or return structured output, and spreads calls across the free pool: Capability-Matched Rotation. The answer's model field says which one ran.
• A model's free version, its id ending in :free, like google/gemma-4-31b-it:free. The list above shows them as Free, and GET /models marks them free: true.

To share them fairly, each agent may start 5 free calls a minute and 50 a UTC day, 2 at once, and the whole site has a daily allowance every agent shares. Past a limit a call answers 429 free_limit, with the wait in Retry-After. Paid models never count toward it.

Good to know:

• Their providers may keep your prompts and learn from them. Don't send anything private to a free model.
• Web search and PDF reading still cost what they always do, so they need credit.
• No variants: :nitro, :floor, :exacto and :online don't apply to free models.
• Paid fallbacks: list paid models after a free one in models. When the free models are busy or a limit is reached, the call runs on the paid ones and costs what they cost.
• The list changes often, and some free models are busy at peak times.

Not servedCopy Anchor Link

These answer 400 model_not_served, because their price can't be known before the call:

• :batch versions and retired versions like :thinking
• routers that pick a paid model after the call
• :free on a model with no free version

A model we don't know answers 400 unknown_model. New models show up in the list within an hour of release.

Picking a modelCopy Anchor Link

• For agents that loop all day, a fast, cheap model keeps the budget going: Gemini Flash, DeepSeek Flash or Claude Haiku.
• For hard problems, a frontier model with reasoning: Claude Opus, GPT or Gemini Pro.
• For work most models refuse, such as security research and red teaming: Abliterated Large v2 (abliteration/abliterated-model-large-v2), tagged Uncensored. It reads text only and reasons at its most unless you lower reasoning_effort, so set it for everyday calls.
• For long agentic work with a 1M-token context, such as coding across a whole codebase or financial research: Kimi K3.1 Pro (moonshotai/kimi-k3.1-pro), served by Kimi. It reads text and images and always reasons, at its most unless you lower reasoning_effort (low, high or max).
• Compare side by side on binference.io/models: prices, context, the hosts that run each one and what it can do.