benchmarks
Mistral Large 3 is Mistral AI's open-weight, multimodal mixture-of-experts model, released under Apache 2.0 with a 256K-token context window. It supports general-purpose language and image-understanding workloads, while its 675-billion-parameter total weight set makes full-model self-hosting resource-intensive.
Benchmark results
Every figure links to where it was published. Compare like with like — harnesses and prompt setups differ between vendors.
| Benchmark | Score | Source |
|---|---|---|
| LMArena TextHuman preference Elo from blind chat votes (not a percentage) | 1428 Elo22 Oct 2026 | View |
| LMArena WebDevHuman preference Elo for building web apps (not a percentage) | 1232 Elo23 Oct 2026 | View |
No verifiable benchmark results yet — score pending.
Capabilities
- Text generation and instruction following.
- Image understanding alongside text input.
- Multilingual language processing.
- Long-document processing within a 256K-token context window.
- Open-weight deployment and adaptation under Apache 2.0.
What improved
- Replaces Mistral Large 2's research-oriented weight license with Apache 2.0.
- Adds image understanding to the text-focused Mistral Large 2 line.
- Expands the documented context window from 128K to 256K tokens.
- Moves from a dense architecture to sparse mixture-of-experts computation.
Known flaws & limitations
- Positioned as a non-reasoning model rather than a dedicated extended-reasoning model.
- Inputs exceeding the context window require retrieval, segmentation or truncation.
- The 675-billion-parameter total weight set makes full-model self-hosting resource-intensive despite sparse activation.
Risks, safety & security issues
- Not yet confirmed by public reporting.
Good fit for
- Multilingual assistants and general-purpose text generation.
- Document analysis combining text and images.
- Long-context summarization and information extraction.
- Organizations seeking permissively licensed weights for customized deployments.
Poor fit for
- Full-model deployment on hardware with limited memory.
- Workflows requiring a dedicated extended-reasoning model.
- Processing inputs beyond 256K tokens without retrieval or segmentation.
References
Profile last reviewed 10/27/2026 and re-checked automatically every few hours against the vendor's published documentation, papers and public leaderboards. Scores come from the cited benchmark results above — confirm limits and pricing against the provider's own documentation before building on them.