Very long inputs
Choose Mistral Large 4
Mistral Large 4 reads up to 1M tokens per request, against 256k for Mistral Large 3.
Mistral Large 3 vs Mistral Large 4: price, context and capabilities
Choose Mistral Large 4 for very long inputs, the hardest reasoning tasks, reasoning over long documents and the latest release; on the other points the two are level or the data is missing.
In short
- A price comparison is not possible: one of them has no comparable published price.
- Mistral Large 4 reads more per request: 1M tokens against 256k.
- Mistral Large 4 is the newer model (October 2026, against December 2025).
Which should you choose?
Each answer below comes straight from the published prices, specifications and independent benchmark results.
Hardest reasoning tasks
Choose Mistral Large 4
Mistral Large 4 scores 38.4 on the Artificial Analysis Intelligence Index, against 9.3 for Mistral Large 3.
Reasoning over long documents
Choose Mistral Large 4
Mistral Large 4 scores 81.3% on Long-context reasoning (AA-LCR), against 36% for Mistral Large 3.
Latest release
Choose Mistral Large 4
Mistral Large 4 is the newer release (October 2026, against December 2025 for Mistral Large 3).
Specifications side by side
| Mistral Large 3 | Mistral Large 4 | |
|---|---|---|
| Provider | Mistral AI | Mistral AI |
| Category | Language models | Language models |
| Status | Legacy | Preview |
| Headline price | No data | $0.75 / 1M tokens |
| Input | No data | $0.50 / 1M tokens |
| Output | No data | $1.50 / 1M tokens |
| Cached input | No data | No data |
| Context window | 256k | 1M |
| Max output | No data | No data |
| Released | December 2025 | October 2026 |
| Knowledge cutoff | No data | No data |
| Licence | Open weights | Open weights |
| Capabilities | Vision, Tool calling, Structured output, Self-hostable | Vision, Tool calling, Structured output, Self-hostable |
| API model id | mistral-large-2512 | mistral-large-4 |
Benchmarks side by side
Leader in bold.
| Benchmark | Mistral Large 3 | Mistral Large 4 |
|---|---|---|
| Intelligence Index | 9.3 | 38.4 |
| SciCode | 36.6% | 54.2% |
| Humanity's Last Exam | 4.2% | 35% |
| MMMU-Pro (visual reasoning) | 55.7% | 76.4% |
| Long-context reasoning (AA-LCR) | 36% | 81.3% |
| Terminal-Bench Hard | 15.9% | No data |
| GPQA Diamond | 68% | No data |
| IFBench (instruction following) | 36.2% | No data |
| τ²-Bench (agentic tool use) | 24.6% | No data |
Independent evaluations by Artificial Analysis, run the same way for every model, at the reasoning setting shown. Percentages unless marked as an index.
Frequently asked questions
Should I choose Mistral Large 3 or Mistral Large 4?
Choose Mistral Large 4 for very long inputs, the hardest reasoning tasks, reasoning over long documents and the latest release; on the other points the two are level or the data is missing.
Which is cheaper, Mistral Large 3 or Mistral Large 4?
A price comparison is not possible: one of them has no comparable published price.
Which has the larger context window, Mistral Large 3 or Mistral Large 4?
Mistral Large 4 reads more per request: 1M tokens against 256k.
What is the difference between Mistral Large 3 and Mistral Large 4?
Mistral Large 3 is made by Mistral AI and Mistral Large 4 by Mistral AI. A price comparison is not possible: one of them has no comparable published price. Mistral Large 4 reads more per request: 1M tokens against 256k. Mistral Large 4 is the newer model (October 2026, against December 2025).
Which is newer, Mistral Large 3 or Mistral Large 4?
Mistral Large 4 was released on October 2026 and Mistral Large 3 on December 2025.