Llama 4 Scout on live video streams
Llama 4 Scout · Meta · Multimodal · Open-weight
Llama 4 Scout is Meta's natively multimodal mixture-of-experts model, pairing 17B active parameters with an industry-leading 10M-token context window. Llama 4 Scout is not currently in Overshoot's live model catalog. When a model is live in the catalog, the workflow is: publish a camera or screen share over WebRTC, reference the latest frame or a recent segment with an ovs:// URL, and stream the answer back through an OpenAI-compatible chat-completions request.
Llama 4 Scout is not currently in the live Overshoot model catalog (verified against the live model catalog on 2026-07-14). Availability changes over time - query GET /v1beta/models for the current list.
- Developer
- Meta
- Parameters
- 109B MoE · 17B active
- Context window
- 10M tokens
- License
- Llama 4 Community
- Released
- Apr 2025
- Inputs
- Text, images, video frames
- Overshoot availability
- Not in live catalogas of 2026-07-14
What Llama 4 Scout is good at
Scout was trained with early-fusion multimodality, so image understanding is native rather than bolted on. It delivers strong document, chart, and general visual question answering at a serving cost closer to a mid-size dense model, because only 17B of its 109B parameters activate per token.
The 10M-token context window is the standout spec. For video work it means an application can carry long conversation history, tool results, and many sampled frames in a single thread without aggressive truncation.
- Document and chart VQA with grounded, citable answers
- Multi-image reasoning across sampled video frames
- Long-running agent threads that accumulate visual context
Llama 4 Scout and Overshoot's streaming workflow
Llama 4 Scout is not currently in Overshoot's live model catalog. For models that are live, the workflow is: create a Stream, publish a LiveKit video track, then send a chat-completions request whose image_url or video_url is an ovs:// reference. Overshoot retains 600 seconds of frame history, so a request can anchor the latest frame, an exact timestamp, or a bounded recent segment.
If Scout were served through the API, its 17B active parameters would suit token-by-token SSE streaming with low time to first token. For any catalog model, thread_id keeps prompt cache hits across repeated queries against the same stream.
How it compares in the Llama line
Scout is the efficiency-focused sibling of Llama 4 Maverick, which activates the same 17B parameters from a much larger expert pool for higher ceiling quality. Teams that need faster, cheaper answers over live video usually start with Scout and step up only when a task demands it. Llama 3.2 Vision remains an option where the older adapter-based architecture or its licensing profile fits better.
Frequently asked questions
Can Llama 4 Scout analyze live video?
Scout was trained with early-fusion multimodality and handles sampled video frames natively. It is not currently in Overshoot's live model catalog, so it cannot be queried against Overshoot streams today. Catalog models read a live WebRTC stream referenced with an ovs:// URL inside a standard chat-completions request, with answers streamed back over SSE.
Is Llama 4 Scout open source?
Scout ships under the Llama 4 Community License. Weights are downloadable and self-hostable with some commercial conditions.
Is Llama 4 Scout available on Overshoot?
Not currently. Scout is not in Overshoot's live model catalog. The catalog changes over time, so check GET /v1beta/models for the current list of models available through the API.
What is the difference between Llama 4 Scout and Maverick?
Both activate 17B parameters per token, but Maverick draws from a much larger expert pool (400B total parameters) for higher quality at higher serving cost, while Scout keeps a 109B pool and a 10M-token context window aimed at efficient long-context work.