MiniMaxMiniMax M3
MiniMax M3 is a MiniMax open-weights reasoning model, released 1 Jun 2026. Its best configuration ranks #203 of 785 on the BenchLeader Index at 56.4 ±4.6, in the upper half. The ± is the point: 273 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 (80) and lowest in agents & tools (40). At $0.525 per million tokens blended it is mid-priced. Output speed of 84 tokens per second puts it slower than most, with a first answer in 25.0 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
- $0.525/M
- $0.300 in · $1.20 out
- Output speed
- 84 tok/s
- measured by Artificial Analysis
- First answer
- 25 s
- first token 0.99 s
- Context
- 1M
- Full answer
- 31 s
- median, reasoning included
- Cost per run
- $0.508
- one full Intelligence Index run
- Released
- 1 Jun 2026
- Overall index56
- Reasoning55
- Coding51
- Agents & tools40
- Maths54
- Knowledge60
- Instruction following80
- Human preference56
- Multimodal60
- Long context67
- Composite48
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 |
|---|---|---|---|---|---|---|
| no reasoning | – | – | 84 tok/s | 25 s | $0.0005 | Maths 35 · Reasoning 59 |
| xhigh | – | – | 84 tok/s | 25 s | $0.0005 | |
| not statedbest | 56.4 | #203 | 84 tok/s | 25 s | $0.0005 | Agents & tools 40 · Coding 51 · Composite 48 · Human preference 56 · Instruction following 80 · Knowledge 60 · Long context 67 · Maths 54 · Multimodal 60 · Reasoning 55 |
How its index has moved
8 Sept 2026 to 23 Oct 2026The 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.
Metadata
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| AA Openness Indexnot in index | 33.3#34 | Artificial Analysis |
Reasoning
| Benchmark | no reasoning | not stated | Source | Trend |
|---|---|---|---|---|
| GPQA Diamond | 81.3%#120 | 90.9%#35 | Epoch AI Benchmarking Hub | |
| SimpleBencheffort not stated | – | 45.8%#52 | SimpleBench | |
| Arena Hard Promptseffort not stated | – | 1463#89 | Arena (LMArena) | |
| LiveBench 2.0 Reasoningnot in indexeffort not stated | – | 74.5%#46 | LiveBench | |
| LiveBench 1.0 Reasoningnot in indexeffort not stated | – | 74.5%#62 | LiveBench | |
| GPQA Diamond (AA)not in indexeffort not stated | – | 92.9%#27 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in indexeffort not stated | – | 39.0%#95 | Artificial Analysis | |
| GPQA Diamond (Vals)not in indexeffort not stated | – | 92.7%#14 | Vals AI | |
| CritPteffort not stated | – | 3.7%#171 | Artificial Analysis | |
| Chess Puzzlesnot in index | 7.0%#127 | 13.0%#115 | Epoch AI Benchmarking Hub | |
| Mystery Game Puzzlesnot in index | 8.0%#119 | 7.0%#123 | Epoch AI Benchmarking Hub | |
| LMCAnot in indexeffort not stated | – | 33.7%#130 | Epoch AI Benchmarking Hub | |
| DTBenchnot in indexeffort not stated | – | 78.9%#137 | Epoch AI Benchmarking Hub | |
| ForecastBenchnot in indexeffort not stated | – | 64.3%#28 | ForecastBench |
Coding
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| SciCode | 47.1%#127 | SciCode | |
| FrontierCode | 14.7%#41 | Cognition | |
| Arena Coding | 1496#87 | Arena (LMArena) | |
| Arena WebDev | 1482#63 | Arena (LMArena) | |
| LiveBench 2.0 Codingnot in index | 70.9%#40 | LiveBench | |
| LiveBench 1.0 Codingnot in index | 68.2%#64 | LiveBench | |
| SciCode (AA)not in index | 47.1%#135 | Artificial Analysis | |
| LiveCodeBench | 82.2%#54 | Vals AI | |
| SWE-bench (Vals)not in index | 75.0%#42 | Vals AI | |
| Code Migrationnot in index | 19.9%#52 | Vals AI | |
| Code Migration v1.1not in index | 20.9%#25 | Vals AI | |
| SWE-bench Multimodalnot in index | 33.3%#19 | Vals AI | |
| Vibe Code Bench 1-100not in index | 9.2%#24 | Vals AI | |
| Vibe Code Bench v1.1not in index | 47.6%#65 | Vals AI | |
| Surface Evolver Benchnot in index | 55.0%#16 | Epoch AI Benchmarking Hub | |
| ALE-Benchnot in index | 640.0#93 | Epoch AI Benchmarking Hub | |
| Terminal-Bench 4.0 (AA)not in index | 2.0%#118 | Artificial Analysis | |
| Terminal-Bench 2.1 (AA)not in index | 65.2%#82 | Artificial Analysis | |
| FrontierCode 1.1 (AA)not in index | 23.0%#61 | Artificial Analysis |
Agents & tools
Maths
| Benchmark | no reasoning | not stated | Source | Trend |
|---|---|---|---|---|
| OTIS Mock AIME | 26.7%#212 | 71.1%#135 | Epoch AI Benchmarking Hub | |
| ProofBencheffort not stated | – | 18.0%#53 | Vals AI | |
| LiveBench Mathematicsnot in indexeffort not stated | – | 77.0%#65 | LiveBench | |
| Arena Mathseffort not stated | – | 1430#105 | Arena (LMArena) |
Knowledge
| Benchmark | xhigh | not stated | Source | Trend |
|---|---|---|---|---|
| LiveBench 2.0 Data Analysisnot in indexeffort not stated | – | 76.2%#26 | LiveBench | |
| LiveBench Researchnot in indexeffort not stated | – | 68.0%#37 | LiveBench | |
| LiveBench 1.0 Data Analysisnot in indexeffort not stated | – | 76.2%#35 | LiveBench | |
| AA-Omniscienceeffort not stated | – | 1.4#124 | Artificial Analysis | |
| MMLU-Proeffort not stated | – | 84.2%#63 | Vals AI | |
| LegalBencheffort not stated | – | 85.4%#21 | Vals AI | |
| CorpFineffort not stated | – | 68.1%#10 | Vals AI | |
| TaxEvaleffort not stated | – | 72.7%#57 | Vals AI | |
| Excel Modeling Benchmarknot in indexeffort not stated | – | 47.8%#55 | Vals AI | |
| Excel Modeling Benchmark v1.1not in indexeffort not stated | – | 59.8%#28 | Vals AI | |
| MedCodenot in indexeffort not stated | – | 46.3%#37 | Vals AI | |
| MedScribenot in indexeffort not stated | – | 87.3%#18 | Vals AI | |
| Public Benefits Bench v1.1not in indexeffort not stated | – | 64.1%#23 | Vals AI | |
| DrugDiscoveryBenchnot in index | 22.8%#19 | – | Scale AI SEAL | |
| AA-Omniscience: accuracynot in indexeffort not stated | – | 16.7%#384 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in indexeffort not stated | – | 81.6%#12 | Artificial Analysis |
Instruction following
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| LiveBench Languagenot in index | 76.8%#44 | LiveBench | |
| LiveBench Instruction Followingnot in index | 57.5%#60 | LiveBench | |
| IFBench | 82.9%#2 | Artificial Analysis | |
| Arena Instruction Followingnot in index | 1434#88 | Arena (LMArena) |
Human preference
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Text | 1440#94 | Arena (LMArena) | |
| EQ-Bench 4 | 1150#19 | EQ-Bench | |
| Arena Creative Writingnot in index | 1404#102 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1452#83 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1451#86 | Arena (LMArena) |
Multimodal
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Vision | 1251#62 | Arena (LMArena) | |
| LiveBench Designnot in index | 66.2%#42 | LiveBench | |
| MMMU-Pro | 78.5%#69 | Artificial Analysis | |
| MMMU-Pro (Vals)not in index | 81.2%#41 | Vals AI | |
| MortgageTaxnot in index | 68.4%#16 | Vals AI | |
| SAGEnot in index | 50.6%#16 | Vals AI | |
| AA-3Dnot in index | 740#20 | Artificial Analysis |
Long context
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| AA-LCR | 83.0%#26 | Artificial Analysis | |
| Arena Documentnot in index | 1433#38 | Arena (LMArena) | |
| MLCRnot in index | 4.4%#151 | Artificial Analysis |
Composite
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Epoch Capabilities Indexnot in index | 146.9#70 | Epoch AI Benchmarking Hub | |
| LiveBench 2.0 | 53.7%#44 | LiveBench | |
| LiveBench 1.0not in index | 67.3%#62 | LiveBench | |
| AA Intelligence Index v4.3.2 | 29.2#130 | Artificial Analysis | |
| Vals Finance Indexnot in index | 48.4%#25 | Vals AI | |
| Vals Multimodal Indexnot in index | 60.0%#18 | Vals AI | |
| AA Cyber Indexnot in index | 22.4#18 | Artificial Analysis |
Where it wins
Benchmarks where this configuration ranks in the top five of every configuration measured.
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 |
|---|---|---|---|---|---|---|
| CoreWeave | 194 tok/s | 0.71 s | $0.230 | $0.960 | 262k | fp4 |
| Parasail | 128 tok/s | 0.58 s | $0.300 | $1.20 | 1.0M | fp8 |
| AtlasCloud | 126 tok/s | 2.40 s | $0.300 | $1.20 | 524k | fp8 |
| ModelRun [by Modular] | 119 tok/s | 0.64 s | $0.300 | $1.30 | 1.0M | fp8 |
| MiniMax | 100 tok/s | 0.91 s | $0.300 | $1.20 | 524k | fp8 |
| Together | 100 tok/s | 0.41 s | $0.300 | $1.20 | 524k | – |
| NovitaAI | 85 tok/s | 1.44 s | $0.300 | $1.20 | 1M | fp8 |
| GMICloud | 56 tok/s | 2.48 s | $0.240 | $0.960 | 1.0M | fp8 |
| StreamLake | 54 tok/s | 5.20 s | $0.300 | $1.20 | 1M | fp8 |
| MARA | 48 tok/s | 1.83 s | $0.600 | $2.40 | 1.0M | – |
| Venice | 15 tok/s | 3.57 s | $0.300 | $1.20 | 524k | fp8 |
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 $0.060 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0005 | $0.0004 | 28.6 s |
| Summarize a 30-page report | 12,000 / 600 | $0.0043 | $0.0022 | 32.1 s |
| Code edit | 6,000 / 1,500 | $0.0036 | $0.0025 | 42.8 s |
| Agentic coding session | 60,000 / 4,000 | $0.023 | $0.012 | 1.2 min |
| Structured extraction | 2,000 / 200 | $0.0008 | $0.0005 | 27.4 s |
See also
Questions people ask
- What are MiniMax M3's benchmark scores?
- As of 24 Oct 2026, MiniMax M3 scores 56.4 ±4.6 on the BenchLeader Index, strongest in instruction following (80) and weakest in agents & tools (40). That rests on 35 independent benchmark results, and every score on this page links to its publisher.
- How does MiniMax M3 rank?
- MiniMax M3 is #203 of 785 ranked configurations on 24 Oct 2026. 273 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 MiniMax M3 cost per 1M tokens?
- As of 24 Oct 2026, MiniMax M3 costs $0.300 per 1M input tokens and $1.20 per 1M output tokens, $0.525 blended (three input tokens per output token).
- When was MiniMax M3 released?
- MiniMax released MiniMax M3 on 1 Jun 2026. BenchLeader has results for it at 3 reasoning-effort settings.
Data as of 24 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “MiniMax M3: benchmarks, pricing, speed and rank”, data as of 24 Oct 2026, https://www.benchleader.com/models/minimax-m3.