Models

Models

Privatemode gives you access to the following models. For pricing and rate limits, see Pricing and Rate limits.

ModelModel IDTypeInputContext / limitEndpoints
DeepSeek-OCR-2 (preview)deepseek-ocr-2OCRText, image8192 tokens/v1/chat/completions
GLM-5.3 (preview)glm-5.3, glm-latestChatText1M tokens/v1/chat/completions, /v1/completions, /v1/messages
GLM-5.3-Flash (preview)glm-5.3-flash, glm-flash-latestChatText, image1M tokens/v1/chat/completions, /v1/completions, /v1/messages
gpt-oss-120bgpt-oss-120bChatText128k tokens/v1/chat/completions, /v1/completions, /v1/messages
Kimi K2.6 (deprecated)kimi-k2.6, kimi-latestChatText, image256k tokens/v1/chat/completions, /v1/completions, /v1/messages
Qwen3-Embedding 4Bqwen3-embedding-4bEmbeddingText32k tokens/v1/embeddings
Voxtral Mini 3Bvoxtral-mini-3bSpeech-to-textAudio50 MB/v1/audio/transcriptions
Whisper large-v3whisper-large-v3Speech-to-textAudio50 MB/v1/audio/transcriptions

All chat models support streaming, tool calling, and structured outputs.

Reasoning

All chat models perform reasoning steps by default. Control it with the reasoning_effort and chat_template_kwargs fields of the request body. Defaults and accepted values depend on the model:

ModelDefaultControl
GLM-5.3, GLM-5.3-Flashreasoning_effort: maxreasoning_effort: low, high, max. Reasoning can't be switched off, any other value including none is mapped to max.
gpt-oss-120breasoning_effort: mediumreasoning_effort: low, medium, high
Kimi K2.6Enabledchat_template_kwargs: {"thinking": false} switches reasoning off

With the OpenAI and Anthropic SDKs, pass chat_template_kwargs via extra_body.

On /v1/chat/completions, the reasoning text is returned in the reasoning field of the assistant message. The deprecated reasoning_content field holds a copy. On /v1/messages, it's returned as thinking content blocks.

GLM-5.3

Use the model ID glm-latest to always route to the latest available GLM model. The model ID glm-5.2 is deprecated and routes to GLM-5.3.

GLM-5.3-Flash

GLM-5.3-Flash is a smaller, faster GLM model. It supports images as input. Follow this guide to use the feature.

Use the model ID glm-flash-latest to always route to the latest available GLM Flash model.

Kimi K2.6

Kimi K2.6 is deprecated and will be removed in a future release. Migrate to GLM-5.3 or GLM-5.3-Flash; use GLM-5.3-Flash for vision tasks.

The model supports images as input.

Use the model ID kimi-latest to always route to the latest available Kimi model.

Qwen3-Embedding 4B

The model uses Matryoshka training and supports output dimensions of 1024 or 2560, set via the dimensions field in the embeddings request. For most tasks, 1024 dimensions is sufficient. For other dimensionalities, truncate the returned vector client-side and re-normalize it afterward.

Voxtral Mini 3B

Use sufficiently high-quality audio with adequate bit rates. Consider Whisper large-v3 if you face issues with low quality audio.

DeepSeek-OCR-2

This OCR model requires special prompt formatting. Follow this guide to use it correctly.