Mistral AIAuto-detectedMagistral Medium 1.2
Magistral Medium 1.2 is a Mistral AI proprietary model, released 18 Sept 2025. Its best configuration ranks #403 of 372 on the BenchLeader Index at 45.2 ±6.0, in the lower half. It scores highest in long context (53) and lowest in knowledge (38). At $2.75 per million tokens blended it is pricier than most ranked models. Last measured 10 Sept 2026.
- Blended price
- $2.75/M
- $2.00 in · $5.00 out
- Output speed
- –
- First answer
- –
- Context
- 40k
- Overall index45
- Coding47
- Agents & tools45
- Maths40
- Knowledge38
- Instruction following45
- Multimodal44
- Long context53
- Composite44
Versions
Mistral AI has shipped 2 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.
| Model | Released | Index | Rank |
|---|---|---|---|
| Magistral Medium 1.2this page | 18 Sept 2025 | 45.2 | #403 |
| Magistral Mediummedium | – | 39.8 | #534 |
Benchmark results
One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | default | Source |
|---|---|---|
| GPQA Diamond (AA)not in index | 73.9%#255 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 10.3%#258 | Artificial Analysis |
| GPQA Diamond (Vals)not in index | 62.4%#103 | Vals AI |
Agents & tools
| Benchmark | default | Source |
|---|---|---|
| Terminal-Bench Hard | 12.9%#200 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 52.0%#191 | Artificial Analysis |
Maths
| Benchmark | default | Source |
|---|---|---|
| AIME (Vals) | 83.5%#42 | Vals AI |
| MGSM | 74.6%#68 | Vals AI |
Knowledge
| Benchmark | default | Source |
|---|---|---|
| AA-Omniscience | -26.7#209 | Artificial Analysis |
| MMLU-Pro | 68.7%#121 | Vals AI |
| LegalBench | 54.8%#131 | Vals AI |
| CorpFin | 47.4%#105 | Vals AI |
| TaxEval | 61.9%#117 | Vals AI |
| MedQA | 89.5%#50 | Vals AI |
Instruction following
| Benchmark | default | Source |
|---|---|---|
| IFBench | 43.0%#225 | Artificial Analysis |
Multimodal
| Benchmark | default | Source |
|---|---|---|
| MMMU-Pro | 59.6%#183 | Artificial Analysis |
Long context
| Benchmark | default | Source |
|---|---|---|
| AA-LCR | 53.0%#247 | Artificial Analysis |
Composite
| Benchmark | default | Source |
|---|---|---|
| AA Intelligence Index | 11.8#280 | Artificial Analysis |
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0023 | – | – |
| Summarise a 30-page report | 12,000 / 600 | $0.027 | – | – |
| Code edit | 6,000 / 1,500 | $0.019 | – | – |
| Agentic coding session | 60,000 / 4,000 | $0.140 | – | – |
| Structured extraction | 2,000 / 200 | $0.0050 | – | – |
See also
Data as of 10 Sept 2026. Compare with another model.