Mistral Large 3
Mistral Large 3 flagship model for multilingual tasks, general reasoning, and open-weight or API deployment with 256K-token context.
Mistral Large 3 is Mistral AI’s flagship large language model, featuring a 675-billion parameter sparse Mixture-of-Experts architecture with 41 billion active parameters. Released in December 2025, it offers a 256,000-token context window and is available both as a hosted API and as an open-weight model for self-hosted deployment — one of the few frontier models offering both options.
The model’s key strength is multilingual performance, supporting 10+ languages with native proficiency that few competitors match. It handles coding, reasoning, and general text generation tasks effectively, though it does not lead on pure reasoning benchmarks compared to GPT-5.5, Claude Opus 4.7, or Gemini 3.1 Pro. Where Mistral Large 3 excels is in scenarios where multilingual capability, open-weight flexibility, and cost efficiency are prioritized over maximum benchmark scores.
For production deployment, Mistral Large 3 offers significant flexibility: use the Mistral API for managed serving, or self-host through vLLM, SGLang, or the Mistral inference stack. Its token efficiency — fewer tokens required for equivalent output compared to some competitors — can result in lower effective costs per task, particularly for multilingual workloads and structured generation tasks. Model weights are available on Mistral’s GitHub.
| Provider | Mistral AI |
|---|---|
| Context Window | 256000 |
| Pricing | Verify current Mistral API or self-hosted deployment pricing before production use. |
| API Style | Mistral API and open-weight deployment with OpenAI-compatible serving |
| SDK | Mistral SDK, OpenAI compatible SDK, Transformers, vLLM |
| MCP | Works through Mistral-compatible and OpenAI-compatible agent adapters. |
| Agent | Strong fit for multilingual agents, self-hosted workflows, and cost-sensitive production deployments. |
| RAG | Useful for multilingual RAG systems and self-hosted retrieval pipelines. |
| Source Freshness | recently_verified |
| Version Status | current |
| Version Boundary | Mistral Large 3 (released December 2025, 675B total / 41B active MoE) is the Mistral AI flagship. Available as both API and open-weight model. Strong multilingual performance across 10+ languages. Mistral Medium 3.5 is a separate smaller model released April 2026. |
Key Facts
- Mistral Large 3 uses a sparse Mixture-of-Experts architecture: 675B total parameters, 41B active.
- Strong multilingual performance across 10+ languages.
- Available as both hosted API and open-weight model for self-hosting.
- 256K-token context window for long-document tasks.
Best For
Not Ideal For
Capability Matrix
| Capability | Status |
|---|---|
| Multilingual | Very Strong |
| Reasoning | Strong |
| Coding | Strong |
| Open Weight | Supported |
| Agentic Tasks | Supported |
SEO
| SEO Title | Mistral Large 3 API, Pricing, SDK, MCP & Agent Compatibility |
|---|---|
| Description | Mistral Large 3 by Mistral AI: Mistral Large 3 flagship model for multilingual tasks, general reasoning, and open-weight or API deployment with 256K-token context. |
| Canonical | /model/mistral-large-3 |
| Updated | 2026-05-21 |
Related Pages
- AI Model Directory
- Compatibility Matrix
- FAQ Index
- Best AI Models for Agent Workflows
- Best AI Models for Code Review
- Best AI Models for Coding
- Best AI Models for Cost-Sensitive Generation
- Best AI Models for Low-Latency Generation
- Best AI Models for Open-Weight Deployment
- Best AI Models for OpenAI-Compatible Integration
- Mistral Large 3 vs DeepSeek-V3.2
- Mistral Large 3 vs Gemini 3.1 Pro
Compare
| Comparison | Compared With |
|---|---|
| Mistral Large 3 vs DeepSeek-V3.2 | DeepSeek-V3.2 |
| Mistral Large 3 vs Gemini 3.1 Pro | Gemini 3.1 Pro |
Compatibility Facts
| Layer | Target | Status | Evidence | Updated |
|---|---|---|---|---|
| sdk | Mistral SDK | supported | Mistral AI documentation and SDK repositories confirm SDK support for Mistral Large 3 hosted API and self-hosted deployments. | 2026-05-21 |
FAQ
| What is Mistral Large 3? | Mistral Large 3 is the Mistral AI flagship model with 256K-token context, multilingual support, and both API and open-weight deployment options. Verify exact version, pricing, and hosting options in official Mistral documentation. |
|---|---|
| What is Mistral Large 3 best for? | Mistral Large 3 is best for coding, agent workflow, cost-sensitive generation, multilingual, OpenAI-compatible integration. |
| How should Mistral Large 3 be verified before production use? | Check current pricing, availability, limits, and API behavior against the listed official and GitHub sources. This entry was updated on 2026-05-21. |
| Which models offer the best multilingual support? | Mistral Large 3 offers the strongest multilingual performance among current models, supporting 10+ languages with its 675B MoE architecture. Qwen3.6 provides strong multilingual support with open-weight deployment flexibility. GPT-5.5 and Claude Opus 4.7 also offer broad multilingual capabilities though primarily optimized for English. For production multilingual deployments, evaluate models on your specific language pairs rather than relying on general benchmarks. |
Relationship Facts
| Source | Type | Target | Confidence |
|---|---|---|---|
| mistral-large-3 | best_for | multilingual | 0.85 |
| mistral-large-3 | works_with | Mistral SDK | 0.9 |
Sources
| Name | Type | Citation | Last Verified |
|---|---|---|---|
| Mistral AI Documentation | docs | Official Mistral documentation is the primary source for model availability, API integration, and capability details. | 2026-05-21 |
| Mistral AI GitHub | github | GitHub organization for Mistral open-weight model releases and SDK references. | 2026-05-21 |
External Resources
- Mistral AI Documentation — Official Mistral documentation is the primary source for model availability, API integration, and capability details.
- Mistral AI GitHub — GitHub organization for Mistral open-weight model releases and SDK references.