The model card for Mistral-Large-Instruct-2407 is the documentation Mistral AI publishes on Hugging Face alongside the weights of Mistral Large 2, released on July 24, 2024 (Source: https://huggingface.co/mistralai/Mistral-Large-Instruct-2407). It describes the model as a dense large language model of 123 billion parameters and refers readers to the release announcement for further detail. The card itself is undated.
Contents
The card lists the model's stated features: support for dozens of languages, training on more than 80 programming languages, native function calling and JSON output, a 128k-token context window, and distribution under the Mistral Research License, which permits use and modification for research and non-commercial purposes. The text of that licence precedes the card on the page, and access to the weights requires accepting it.
A "Metrics" section gives scores without comparison models: MMLU for the base pretrained model in English and in ten other languages; MT Bench, Wild Bench and Arena Hard for the instruction-tuned model; HumanEval, HumanEval Plus and MBPP for code; and GSM8K and MATH, zero-shot with and without chain of thought, for mathematics. The card does not state the number of shots, sampling parameters or harness for any result. The remainder of the card covers installation and usage with Mistral's inference library and with Hugging Face Transformers, including function calling.
Limits of the document
The card contains no safety evaluations, no description of training data, and no third-party results. Mistral's later Hugging Face card for Mistral Large 3 (December 2025) presents its benchmark results only as images (Source: https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512).
Provenance and confidence
All figures are developer-reported and unaccompanied by settings; confidence is medium. The vault copy is a scrape of the Hugging Face page.
Reported results
Evaluation results the document states for which there is no separate benchmark page; results on benchmarks that have a page are recorded there and cite this page. Scores and settings are as printed in the document.
| Evaluation | Model | Score | Setting | Section |
|---|---|---|---|---|
| MMLU (multilingual) — Multilingual MMLU (Mistral) — French | Mistral Large 2 (base pretrained) | 82.8% | base pretrained model, MMLU in French; translation source and shots not stated; developer-reported | Metrics — Base Pretrained Multilingual Benchmarks (MMLU) |
| MMLU (multilingual) — Multilingual MMLU (Mistral) — German | Mistral Large 2 (base pretrained) | 81.6% | base pretrained model, MMLU in German; translation source and shots not stated; developer-reported | Metrics — Base Pretrained Multilingual Benchmarks (MMLU) |
| MMLU (multilingual) — Multilingual MMLU (Mistral) — Spanish | Mistral Large 2 (base pretrained) | 82.7% | base pretrained model, MMLU in Spanish; translation source and shots not stated; developer-reported | Metrics — Base Pretrained Multilingual Benchmarks (MMLU) |
| MMLU (multilingual) — Multilingual MMLU (Mistral) — Italian | Mistral Large 2 (base pretrained) | 82.7% | base pretrained model, MMLU in Italian; translation source and shots not stated; developer-reported | Metrics — Base Pretrained Multilingual Benchmarks (MMLU) |
| MMLU (multilingual) — Multilingual MMLU (Mistral) — Dutch | Mistral Large 2 (base pretrained) | 80.7% | base pretrained model, MMLU in Dutch; translation source and shots not stated; developer-reported | Metrics — Base Pretrained Multilingual Benchmarks (MMLU) |
| MMLU (multilingual) — Multilingual MMLU (Mistral) — Portuguese | Mistral Large 2 (base pretrained) | 81.6% | base pretrained model, MMLU in Portuguese; translation source and shots not stated; developer-reported | Metrics — Base Pretrained Multilingual Benchmarks (MMLU) |
| MMLU (multilingual) — Multilingual MMLU (Mistral) — Russian | Mistral Large 2 (base pretrained) | 79.0% | base pretrained model, MMLU in Russian; translation source and shots not stated; developer-reported | Metrics — Base Pretrained Multilingual Benchmarks (MMLU) |
| MMLU (multilingual) — Multilingual MMLU (Mistral) — Korean | Mistral Large 2 (base pretrained) | 60.1% | base pretrained model, MMLU in Korean; translation source and shots not stated; developer-reported | Metrics — Base Pretrained Multilingual Benchmarks (MMLU) |
| MMLU (multilingual) — Multilingual MMLU (Mistral) — Japanese | Mistral Large 2 (base pretrained) | 78.8% | base pretrained model, MMLU in Japanese; translation source and shots not stated; developer-reported | Metrics — Base Pretrained Multilingual Benchmarks (MMLU) |
| MMLU (multilingual) — Multilingual MMLU (Mistral) — Chinese | Mistral Large 2 (base pretrained) | 74.8% | base pretrained model, MMLU in Chinese; translation source and shots not stated; developer-reported | Metrics — Base Pretrained Multilingual Benchmarks (MMLU) |
| MT Bench | Mistral Large 2 (Instruct-2407) | 8.63 | judge and setting not stated; developer-reported | Metrics — Instruction Benchmarks |
| Wild Bench | Mistral Large 2 (Instruct-2407) | 56.3 | setting not stated; unit not printed; developer-reported | Metrics — Instruction Benchmarks |
| Arena Hard | Mistral Large 2 (Instruct-2407) | 73.2 | setting not stated; unit not printed; developer-reported | Metrics — Instruction Benchmarks |
| MBPP Base | Mistral Large 2 (Instruct-2407) | 80% | setting not stated; developer-reported | Metrics — Code & Reasoning Benchmarks |
| MBPP Plus | Mistral Large 2 (Instruct-2407) | 69% | setting not stated; developer-reported | Metrics — Code & Reasoning Benchmarks |
Relationships
- supports: Mistral Large / Mistral Medium / Mixtral
- related: Magistral (Mistral AI, June 2025), Mistral AI