Overview
Mistral Large 2.1, released by Mistral AI on November 18, 2024 and served as mistral-large-2411, was the company's top-tier general-purpose large language model. It is an incremental refresh of Mistral Large 2 (version 24.07): the same 123-billion-parameter dense transformer, re-tuned for better long-context handling, more reliable function calling, and stronger support for system prompts. Mistral positioned it for high-complexity tasks like agentic workflows with strict JSON output, retrieval-augmented generation over large contexts, and code generation across 80+ programming languages.
The model ships with a 128k-token context window and is multilingual, covering English, French, German, Spanish, Italian, Portuguese, Chinese, Japanese, Russian, and Korean among others. Unlike its multimodal sibling Pixtral Large 2411 (which adds vision), Mistral Large 2.1 is a text-only model. Its weights were published on Hugging Face as Mistral-Large-Instruct-2411 under the Mistral Research License, making it open-weight for research and non-commercial use while requiring a separate commercial license for production deployment.
Mistral Large 2.1 is now a legacy model. Mistral AI scheduled it for deprecation on its hosted API (La Plateforme) on February 27, 2026, pointing users toward newer models such as Mistral Medium 3.5, and it has also been retired or deprecated on third-party platforms including Azure AI Foundry and GitHub Models. The open weights remain downloadable for self-hosting.
| Released | 2024-11-18 |
|---|---|
| License | Mistral Research License (MRL 0.1) — free for research and non-commercial use; commercial deployment requires a separate Mistral Commercial License. |
| Weights | Open weights |
| Parameters | 123B (dense; 122.6B in the released safetensors) |
| Context | 128K tokens |
| Max output | Not published by Mistral |
| Architecture | Dense decoder-only transformer (MistralForCausalLM), 123B parameters, BF16 weights. Not a mixture-of-experts model. Same 123B base as Mistral Large 2 (24.07), with an updated instruct/post-training pass. |
| Knowledge cutoff | July 2024 (per OpenRouter; not officially published by Mistral) |
| Modalities | text |
| Status | Deprecated / legacy. Mistral AI scheduled mistral-large-2411 for deprecation on its La Plateforme API on 2026-02-27, with Mistral Medium 3.5 as the suggested replacement. It was also retired on Azure AI Foundry (effective 2026-01-30) and deprecated in GitHub Models (2025-04-16). Open weights remain available on Hugging Face. |
Benchmarks
Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.
Pricing
| Input | $2.00 / 1M tokens per 1M tokens |
|---|---|
| Output | $6.00 / 1M tokens per 1M tokens |
Mistral's own pricing page no longer lists this legacy model; the $2 input / $6 output rate is the widely reported API price as carried by OpenRouter and other providers.
Strengths
- 123B dense transformer with strong reasoning, knowledge, and coding across 80+ programming languages
- 128k-token context window with improved long-context adherence versus the 24.07 release
- Native function calling and JSON-structured output tuned for agentic workflows
- Stronger system-prompt support than earlier Mistral Large versions
- Open weights on Hugging Face under the Mistral Research License for research/self-hosting
- Multilingual coverage across major European and Asian languages
Best for
- Agentic pipelines that need precise instruction following and reliable JSON output
- Retrieval-augmented generation (RAG) over large documents needing strong context adherence
- Code generation and refactoring across many programming languages
- Multilingual chat, summarization, and content generation
- On-premise / self-hosted deployment for research where open weights are required
How to access
| Provider | Model ID |
|---|---|
| Mistral AI (La Plateforme) ↗ | mistral-large-2411 |
| OpenRouter ↗ | mistralai/mistral-large-2411 |
| Google Cloud Vertex AI ↗ | mistral-large-2411 |
| Hugging Face (weights) ↗ | mistralai/Mistral-Large-Instruct-2411 |
Mistral Large — every version
The full lineage of the Mistral Large line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
| Version | Released | Context | License |
|---|---|---|---|
| Mistral Large 3current | 2025-12-02 | 256K | Apache-2.0 |
| Mistral Large 2.1 (24.11) | 2024-11-18 | — | Open weights |
| Mistral Large 2 (24.07) | 2024-07-24 | — | Open weights |
| Mistral Large (24.02) | 2024-02-26 | — | Proprietary |
FAQ
Is Mistral Large 2.1 the same as Mistral Large 2 (24.07)?
It is the November 2024 refresh of the same model line. Both are 123B-parameter dense models; version 24.11 (mistral-large-2411) keeps the same base and adds improvements to long-context handling, function calling, and system-prompt support over the July 2024 release (mistral-large-2407).
Is Mistral Large 2.1 open source?
Its weights are open and downloadable on Hugging Face as Mistral-Large-Instruct-2411, but under the Mistral Research License, which permits research and non-commercial use only. Commercial deployment requires a separate Mistral Commercial License, so it is open-weight rather than fully open-source.
Is Mistral Large 2.1 still available?
It is now a legacy model. Mistral scheduled mistral-large-2411 for deprecation on its hosted API on February 27, 2026, suggesting Mistral Medium 3.5 as a replacement, and it has been retired or deprecated on platforms such as Azure AI Foundry and GitHub Models. The open weights remain available for self-hosting.
Does Mistral Large 2.1 support images?
No. Mistral Large 2.1 is a text-only model. The multimodal vision counterpart released at the same time is Pixtral Large 2411, a separate model.