Pixtral Large on live WebRTC streams

What Pixtral Large is good at

Pixtral Large pairs the reasoning strength of the Mistral Large 2 text model with the same native-resolution vision encoder used in Pixtral 12B, so it inherits both the deep reasoning of a frontier text model and careful handling of small visual detail. It is built to sit alongside the largest closed multimodal models on complex visual reasoning tasks rather than compete only with other open-weight models.

That combination makes it a strong choice for tasks that mix visual grounding with multi-step reasoning: interpreting a technical diagram and explaining its implications, or answering a question that requires connecting several details across a frame.

• Complex visual reasoning that combines grounding with multi-step logic
• Detailed document and diagram interpretation
• Workloads that want frontier-level quality with open weights

Pixtral Large and live WebRTC streams

Pixtral Large is not currently in Overshoot's live model catalog; GET /v1beta/models lists what is live at any given time. Live footage on Overshoot works through one mechanism for every model in the catalog: publish a camera or screen share over WebRTC to a Stream, which keeps 600 seconds of frame history available, then reference that footage in a chat-completions request with an ovs:// URL, anchored to the latest frame, an exact timestamp, or a recent segment.

For a model as large as Pixtral Large, SSE streaming matters: an application sees tokens as they are generated rather than waiting on the full response, and a consistent thread_id keeps a prompt cache warm across repeated questions against the same stream, which helps offset the cost of running a model this size wherever it is served.

Pixtral Large versus Pixtral 12B

Pixtral Large and Pixtral 12B share a vision encoder design, but Large’s 123B parameter text backbone gives it much deeper reasoning at higher serving cost. Teams typically default to Pixtral 12B for routine OCR and chart reading, and reach for Pixtral Large when a question demands frontier-level reasoning about what is visible on the stream, not just accurate extraction.

Frequently asked questions

Can Pixtral Large analyze live video?

Pixtral Large accepts images and video frames, so it can analyze video by reading sampled frames when self-hosted or served under license. It is not currently in Overshoot's live model catalog, so it cannot be referenced against an Overshoot WebRTC stream today. Check GET /v1beta/models for the current catalog.

Is Pixtral Large open source?

Pixtral Large ships under the Mistral Research License, with downloadable weights for research use and a separate commercial license available from Mistral AI for production deployments.

Is Pixtral Large available on Overshoot?

Not currently. Pixtral Large is not in Overshoot's live model catalog. The catalog changes over time, so check GET /v1beta/models for the up-to-date list of available models.

When should I use Pixtral Large instead of Pixtral 12B?

Reach for Pixtral Large when a task needs frontier-level visual reasoning, such as interpreting a complex diagram or combining several details into a conclusion. For routine OCR, chart reading, or high-throughput workloads, Pixtral 12B is cheaper and fast enough.

Related models

• Pixtral 12B
• Mistral Small 3.1
• Llama 4 Maverick
• Qwen3-VL 235B-A22B