Primary use stated or inferred cautiously from official documentation.
M
MISTRAL AI / mistral-large-4
Mistral Large 4
A multimodal MoE model with approximately one trillion parameters and 52B active including embeddings and output layers. It accepts text and images and produces text. FP8 and NVFP4 weights were published on October 29; the API had been in preview since October 6.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | mistral-large-4 |
|---|---|
| Model type | Text and reasoning · Code · Vision and documents |
| Access model | Open weights and paid API |
| License | Mistral Ecosystem Open-Weight License |
| Deployment | Self-hosting or hosted API |
| Release | 29 OCT 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text · Image |
| Exit | Text |
| Context window | 1M tokens |
| maximum output | Not published |
| Reasoning | low / medium / high |
| Published tools | Features |
| Structured Outings | JSON Schema |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face · Mistral API · vLLM · SGLang · Transformers |
Coding with tools, document analysis and agents with long context, through the API or self-hosted infrastructure.
The checkpoint is mistralai/Mistral-Large-4-1T-A52B; the API uses mistral-large-4. The 49B active count excludes layers included in the 52B count. Results and memory requirements are not transferred between FP8 and NVFP4.
03 / CAPABILITIES
What it can do
Orientation
Coding with tools, document analysis and agents with long context, through the API or self-hosted infrastructure.
Context
1M tokens · Not published
Tools and integration
Features
Access
Hugging Face and Mistral API
Input modalities
TextImage
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | 0.70 USD / 1M tokens | Standard API rate |
| Cached input | 0.07 USD / 1M tokens | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | 2.10 USD / 1M tokens | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
Standard API rates in USD per million tokens, verified on October 30. Self-hosting has additional costs. The weights license modifies MIT and requires a separate commercial license to offer Model-as-a-Service.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
Mistral reports Artificial Analysis results for coding and general intelligence. These are different indices, not universal accuracy. We retain the source and checkpoint; Inferama has not reproduced the evaluations.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| Artificial Analysis Coding Agent Index | 48 % | Published score | View source ↗ |
| Artificial Analysis Intelligence Index v5 | 26 | Published score | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
Mistral Large 4
API preview.
Mistral Large 4
FP8 and NVFP4 weights released.
07 / SOURCES