# Models and names

Which models we serve, the names they answer to, and how a model gets added or removed.

## The catalog

75 models, reachable under 137 names, as announced by our seller on 20 Nov 2026. The live, dated table with prices, context sizes and capabilities is on [Use](/use/#models). It is the source; this page explains what is in it.

Families today: Claude, GPT, Grok, Gemini, Kimi, GLM, DeepSeek, Qwen, MiniMax and a few more. Text only: no image generation, no audio.

## Names and aliases

AntSeed buyers route by model name. Most models answer to more than one name, so that the name your tool already uses works without a mapping:

- the popular name, as people type it: `kimi-k3`, `claude-opus-5-5`;
- the canonical name Venice uses;
- for some models, a capitalised variant, because some tools send it that way.

All names of one model cost the same. The table on Use lists every name under its model.

## Long-context names

Some models have a second name ending in `-1m` or similar. It is the same model with the long-context tier enabled. Venice prices long context higher (for some models, several times higher past a threshold), and so do we; the table shows both prices. If your requests stay short, use the plain name.

## Capabilities

The table marks, per model: vision (image input), tool calling, and context size. We serve what Venice serves for that model; if a request asks for a capability the model does not have, Venice returns an error and we pass it through unchanged.

Venice labels some models "private" (served without retaining data) and some "anonymized" (passed to the model's maker with your identity stripped). Those labels are Venice's, and the table shows them as Venice publishes them.

## Parameters we pass through

Standard chat-completions fields: `messages`, `max_tokens`, `temperature`, `top_p`, `stream`, `tools`, `tool_choice`, `response_format`, `stop`, and the reasoning controls the model supports. Venice rejects unknown parameters with a 400; we strip the ones we know Venice refuses and pass the rest. If a request still comes back 400, the body names the parameter.

Streaming works, including tool calls inside a stream.

## When a model is added

New Venice models usually appear here on the day Venice announces them. Before a model is sold, a probe checks that it answers, identifies itself as the model it claims to be, and follows the parameters we rely on. If it fails the probe it is not listed.

## When a model is removed

If Venice retires a model, or it stops answering the probe, it leaves the catalog. A request for a name that is not in the catalog gets a 404 from the proxy before it reaches us.

## What the table does not say

- It does not rank models. Which one fits your work is yours to judge.
- It does not show the price you will pay at a given hour; it shows the dated snapshot. See [Prices through the day](/docs/prices/).

_Last updated 2026-10-20._
