Mistral AIMistral Medium 3.5
Mistral Medium 3.5 is a Mistral AI open-weights reasoning model, released 28 Apr 2026. Its best configuration ranks #397 of 785 on the BenchLeader Index at 49.4 ±6.2, in the lower half. The ± is the point: 370 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in instruction following (68) and lowest in coding (37). At $3.00 per million tokens blended it is pricier than most ranked models. Output speed of 159 tokens per second puts it in the fastest quarter, with a first answer in 14.8 s. It has been measured at 3 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 22 Oct 2026.
- Blended price
- $3.00/M
- $1.50 in · $7.50 out
- Output speed
- 159 tok/s
- measured by Artificial Analysis
- First answer
- 15 s
- first token 0.55 s
- Context
- 262k
- Full answer
- 18 s
- median, reasoning included
- Cost per run
- $0.502
- one full Intelligence Index run
- Released
- 28 Apr 2026
- Overall index49
- Reasoning45
- Coding37
- Agents & tools44
- Maths52
- Knowledge45
- Instruction following68
- Human preference47
- Multimodal51
- Long context60
- Composite45
Versions
Mistral AI has shipped 4 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 |
|---|---|---|---|
| Mistral Medium 3.5this page | 28 Apr 2026 | 49.4 | #397 |
| Mistral Medium 3.1 | 12 Aug 2025 | 46.1 | #494 |
| Mistral Medium 3 | 7 May 2025 | 44.1 | #546 |
| Mistral Medium 2312 | 11 Dec 2023 | – | – |
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| medium | 41.1 | #639 | 159 tok/s | 15 s | $0.0029 | Coding 37 · Composite 36 · Human preference 36 · Knowledge 39 · Maths 38 · Reasoning 42 |
| high | – | – | 159 tok/s | 15 s | $0.0029 | Agents & tools 29 · Knowledge 45 |
| not statedbest | 49.4 | #397 | 159 tok/s | 15 s | $0.0029 | Agents & tools 44 · Coding 37 · Composite 45 · Human preference 47 · Instruction following 68 · Knowledge 45 · Long context 60 · Maths 52 · Multimodal 51 · Reasoning 45 |
How its index has moved
8 Sept 2026 to 23 Oct 2026- not stated
- medium
The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
Benchmark results
See all LLM benchmarks. 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 | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| Arena Hard Prompts | 1236#297 | – | 1445#117 | Arena (LMArena) | |
| GPQA Diamond (AA)not in index | 35.0%#494 | – | 74.8%#254 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 3.5%#592 | – | 13.8%#267 | Artificial Analysis | |
| GPQA Diamond (Vals)not in index | – | 34.9%#126 | – | Vals AI | |
| Kagi LLM Benchmark | 50.0%#84 | – | 41.4%#102 | Kagi LLM Benchmark | |
| CritPteffort not stated | – | – | 0.0%#315 | Artificial Analysis | |
| LMCAnot in indexeffort not stated | – | – | 26.1%#147 | Epoch AI Benchmarking Hub | |
| DTBenchnot in indexeffort not stated | – | – | 75.5%#148 | Epoch AI Benchmarking Hub |
Coding
| Benchmark | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| SciCodeeffort not stated | – | – | 40.2%#161 | SciCode | |
| WeirdMLeffort not stated | – | – | 43.7%#87 | WeirdML | |
| FrontierCodeeffort not stated | – | – | 8.0%#45 | Cognition | |
| Arena Coding | 1263#299 | – | 1478#109 | Arena (LMArena) | |
| Arena WebDeveffort not stated | – | – | 1264#122 | Arena (LMArena) | |
| SciCode (AA)not in indexeffort not stated | – | – | 40.2%#172 | Artificial Analysis | |
| SWE-bench (Vals)not in index | – | 66.4%#69 | – | Vals AI | |
| Code Migrationnot in index | – | 5.1%#68 | – | Vals AI | |
| Vibe Code Bench v1.1not in index | – | 2.9%#108 | – | Vals AI | |
| Surface Evolver Benchnot in indexeffort not stated | – | – | 26.9%#26 | Epoch AI Benchmarking Hub | |
| ALE-Benchnot in indexeffort not stated | – | – | 764.0#81 | Epoch AI Benchmarking Hub | |
| Terminal-Bench 4.0 (AA)not in indexeffort not stated | – | – | 0.0%#152 | Artificial Analysis | |
| Terminal-Bench 2.1 (AA)not in indexeffort not stated | – | – | 50.6%#119 | Artificial Analysis |
Agents & tools
| Benchmark | high | not stated | Source | Trend |
|---|---|---|---|---|
| Arena Agenteffort not stated | – | -11.8#47 | Arena (LMArena) | |
| Terminal-Bench Hardeffort not stated | – | 33.3%#92 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in indexeffort not stated | – | 94.2%#33 | Artificial Analysis | |
| Terminal-Bench 2.1 (Vals) | 39.0%#67 | – | Vals AI | |
| GDPval-AA v2.1not in indexeffort not stated | – | 13.0%#109 | Artificial Analysis | |
| τ³-Banking (AA)effort not stated | – | 15.1%#119 | Artificial Analysis | |
| Analyst Agent (AA)not in indexeffort not stated | – | 12.5%#42 | Artificial Analysis | |
| Finance Agent v2not in index | 32.1%#67 | – | Vals AI | |
| Harvey's Legal Agent Benchmark v1not in index | 0.4%#58 | – | Vals AI | |
| Legal Research Benchnot in index | 9.1%#73 | – | Vals AI | |
| Tax Agent Benchnot in index | 4.5%#68 | – | Vals AI | |
| Terminal-Bench 2.0 (Vals)not in index | 30.3%#43 | – | Vals AI | |
| AutomationBenchnot in indexeffort not stated | – | 6.3%#161 | Artificial Analysis | |
| GDP.pdf (AA)not in indexeffort not stated | – | 2.8%#183 | Artificial Analysis | |
| EnterpriseOps-Gymnot in indexeffort not stated | – | 33.7%#36 | Artificial Analysis |
Maths
| Benchmark | medium | not stated | Source | Trend |
|---|---|---|---|---|
| ProofBencheffort not stated | – | 9.0%#66 | Vals AI | |
| Arena Maths | 1220#284 | 1429#107 | Arena (LMArena) |
Knowledge
| Benchmark | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| AA-Omniscienceeffort not stated | – | – | -36.8#309 | Artificial Analysis | |
| MMLU-Pro | – | 75.3%#104 | – | Vals AI | |
| CorpFin | – | 58.8%#80 | – | Vals AI | |
| TaxEval | – | 68.0%#99 | – | Vals AI | |
| PRBench Finance | 39.4%#26 | – | – | Scale AI SEAL | |
| PRBench Legal | 39.5%#33 | – | – | Scale AI SEAL | |
| CaseLaw v2not in index | – | 44.2%#51 | – | Vals AI | |
| Excel Modeling Benchmarknot in index | – | 11.2%#71 | – | Vals AI | |
| MedCodenot in index | – | 33.8%#82 | – | Vals AI | |
| MedScribenot in index | – | 67.7%#95 | – | Vals AI | |
| AA-Omniscience: accuracynot in indexeffort not stated | – | – | 24.7%#239 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in indexeffort not stated | – | – | 18.4%#299 | Artificial Analysis |
Instruction following
| Benchmark | medium | not stated | Source | Trend |
|---|---|---|---|---|
| IFBencheffort not stated | – | 68.8%#78 | Artificial Analysis | |
| Arena Instruction Followingnot in index | 1211#298 | 1421#107 | Arena (LMArena) |
Human preference
| Benchmark | medium | not stated | Source | Trend |
|---|---|---|---|---|
| Arena Text | 1223#302 | 1426#113 | Arena (LMArena) | |
| EQ-Bench 4effort not stated | – | 993#27 | EQ-Bench | |
| Arena Creative Writingnot in index | 1193#298 | 1400#107 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1196#301 | 1432#109 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1221#302 | 1429#123 | Arena (LMArena) |
Multimodal
| Benchmark | high | not stated | Source | Trend |
|---|---|---|---|---|
| Arena Visioneffort not stated | – | 1223#80 | Arena (LMArena) | |
| MMMU-Proeffort not stated | – | 64.9%#180 | Artificial Analysis | |
| MortgageTaxnot in index | 28.9%#90 | – | Vals AI | |
| SAGEnot in index | 37.6%#64 | – | Vals AI |
Long context
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| AA-LCR | 69.3%#226 | Artificial Analysis | |
| MLCRnot in index | 10.0%#113 | Artificial Analysis |
Composite
| Benchmark | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in indexeffort not stated | – | – | 141.4#107 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index v4.3.2 | 5.5#577 | – | 14.2#298 | Artificial Analysis | |
| Vals Multimodal Indexnot in index | – | 34.8%#31 | – | Vals AI |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| Mistral | 145 tok/s | 0.78 s | $1.65 | $8.25 | 262k | – |
| Mistral | 144 tok/s | 0.55 s | $1.50 | $7.50 | 262k | – |
| Mistral | 122 tok/s | 0.51 s | $1.50 | $7.50 | 262k | – |
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 at $0.150 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0029 | $0.0024 | 16.7 s |
| Summarize a 30-page report | 12,000 / 600 | $0.022 | $0.010 | 18.6 s |
| Code edit | 6,000 / 1,500 | $0.020 | $0.014 | 24.3 s |
| Agentic coding session | 60,000 / 4,000 | $0.120 | $0.059 | 40.0 s |
| Structured extraction | 2,000 / 200 | $0.0045 | $0.0025 | 16.1 s |
More from Mistral AI
See also
Questions people ask
- What are Mistral Medium 3.5's benchmark scores?
- As of 24 Oct 2026, Mistral Medium 3.5 scores 49.4 ±6.2 on the BenchLeader Index, strongest in instruction following (68) and weakest in coding (37). That rests on 21 independent benchmark results, and every score on this page links to its publisher.
- How does Mistral Medium 3.5 rank?
- Mistral Medium 3.5 is #397 of 785 ranked configurations on 24 Oct 2026. 370 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
- How much does Mistral Medium 3.5 cost per 1M tokens?
- As of 24 Oct 2026, Mistral Medium 3.5 costs $1.50 per 1M input tokens and $7.50 per 1M output tokens, $3.00 blended (three input tokens per output token).
- When was Mistral Medium 3.5 released?
- Mistral AI released Mistral Medium 3.5 on 28 Apr 2026. BenchLeader has results for it at 3 reasoning-effort settings.
Data as of 24 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “Mistral Medium 3.5: benchmarks, pricing, speed and rank”, data as of 24 Oct 2026, https://www.benchleader.com/models/mistral-medium-3-5.