MiniMaxMiniMax M2.5
MiniMax M2.5 is a MiniMax open-weights reasoning model, released 12 Feb 2026. Its best configuration ranks #280 of 785 on the BenchLeader Index at 53.6 ±4.3, in the upper half. The ± is the point: 297 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 (70) and lowest in reasoning (45). At $0.525 per million tokens blended it is mid-priced. Output speed of 65 tokens per second puts it slower than most, with a first answer in 32.2 s. It has been measured at 2 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 23 Oct 2026.
- Blended price
- $0.525/M
- $0.300 in · $1.20 out
- Output speed
- 65 tok/s
- measured by Artificial Analysis
- First answer
- 32 s
- first token 1.03 s
- Context
- 205k
- Full answer
- 40 s
- median, reasoning included
- Cost per run
- –
- Released
- 12 Feb 2026
- Overall index54
- Reasoning45
- Coding51
- Agents & tools46
- Maths52
- Knowledge50
- Instruction following70
- Human preference56
- Long context62
- Composite55
Versions
MiniMax has shipped 3 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 |
|---|---|---|---|
| MiniMax M2.7 | 18 Mar 2026 | 55.6 | #225 |
| MiniMax M2.5this page | 12 Feb 2026 | 53.6 | #280 |
| MiniMax-M2.1 | 23 Dec 2025 | 53.8 | #268 |
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 |
|---|---|---|---|---|---|---|
| high | – | – | 65 tok/s | 32 s | $0.0005 | Coding 66 |
| not statedbest | 53.6 | #280 | 65 tok/s | 32 s | $0.0005 | Agents & tools 46 · Coding 51 · Composite 55 · Human preference 56 · Instruction following 70 · Knowledge 50 · Long context 62 · Maths 52 · Reasoning 45 |
How its index has moved
8 Sept 2026 to 21 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.
Reasoning
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Hard Prompts | 1418#154 | Arena (LMArena) | |
| GPQA Diamond (AA)not in index | 84.8%#132 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 20.5%#213 | Artificial Analysis | |
| GPQA Diamond (Vals)not in index | 82.1%#60 | Vals AI | |
| Kagi LLM Benchmark | 55.2%#64 | Kagi LLM Benchmark | |
| ARC-AGI-1 | 63.7%#155 | ARC Prize | |
| ARC-AGI-2 | 4.9%#177 | ARC Prize | |
| CritPt | 1.1%#223 | Artificial Analysis | |
| CL-benchnot in index | 11.4%#21 | Epoch AI Benchmarking Hub | |
| CL-bench Lifenot in index | 6.3%#16 | Epoch AI Benchmarking Hub |
Coding
| Benchmark | high | not stated | Source | Trend |
|---|---|---|---|---|
| Arena Codingeffort not stated | – | 1446#149 | Arena (LMArena) | |
| Arena WebDeveffort not stated | – | 1386#98 | Arena (LMArena) | |
| LiveCodeBencheffort not stated | – | 79.2%#70 | Vals AI | |
| SWE-bench (Vals)not in indexeffort not stated | – | 74.2%#47 | Vals AI | |
| SWE-bench Verified (bash only) | 75.8%#2 | – | SWE-bench | |
| SWE-bench Verified (any scaffold)not in index | 75.8%#4 | – | SWE-bench | |
| Vibe Code Bench v1.1not in indexeffort not stated | – | 14.8%#90 | Vals AI | |
| SWE Atlas: Codebase QnAnot in indexeffort not stated | – | 10.3%#21 | Scale AI SEAL | |
| SWE Atlas: Refactoringnot in indexeffort not stated | – | 19.5%#16 | Scale AI SEAL | |
| SWE Atlas: Test Writingnot in indexeffort not stated | – | 18.6%#22 | Scale AI SEAL | |
| ALE-Benchnot in indexeffort not stated | – | 618.2#96 | Epoch AI Benchmarking Hub |
Agents & tools
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Terminal-Bench | 42.7%#46 | Terminal-Bench | |
| Terminal-Bench Hard | 34.9%#82 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 95.3%#23 | Artificial Analysis | |
| HiL-Bench | 6.3%#29 | Scale AI SEAL | |
| Terminal-Bench 2.0 (Vals)not in index | 41.6%#29 | Vals AI | |
| Vending-Bench 2not in index | -23.2#68 | Andon Labs |
Maths
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| ProofBench | 4.0%#72 | Vals AI | |
| AIME (Vals) | 88.8%#28 | Vals AI | |
| Arena Maths | 1393#156 | Arena (LMArena) |
Knowledge
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| AA-Omniscience | -38.9#318 | Artificial Analysis | |
| MMLU-Pro | 80.1%#85 | Vals AI | |
| LegalBench | 80.0%#89 | Vals AI | |
| CorpFin | 59.6%#74 | Vals AI | |
| TaxEval | 68.2%#96 | Vals AI | |
| MedQA | 92.5%#29 | Vals AI | |
| CaseLaw v2not in index | 53.5%#39 | Vals AI | |
| AA-Omniscience: accuracynot in index | 26.2%#218 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 11.9%#386 | Artificial Analysis |
Instruction following
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| IFBench | 71.6%#52 | Artificial Analysis | |
| Arena Instruction Followingnot in index | 1381#161 | Arena (LMArena) |
Human preference
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Text | 1391#162 | Arena (LMArena) | |
| Arena Creative Writingnot in index | 1357#162 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1396#157 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1405#152 | Arena (LMArena) |
Long context
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| AA-LCR | 73.3%#178 | Artificial Analysis |
Composite
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Epoch Capabilities Indexnot in index | 146.6#74 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index v4.3.2 | 22.8#196 | Artificial Analysis |
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 |
|---|---|---|---|---|---|---|
| StreamLake | 91 tok/s | 1.03 s | $0.270 | $1.08 | 200k | – |
| AtlasCloud | 71 tok/s | 1.60 s | $0.295 | $1.20 | 197k | fp8 |
| DigitalOcean | 64 tok/s | 0.77 s | $0.300 | $1.20 | 66k | – |
| NovitaAI | 44 tok/s | 1.04 s | $0.300 | $1.20 | 205k | fp8 |
| MiniMax | 44 tok/s | 1.23 s | $0.300 | $1.20 | 205k | fp8 |
| MiniMax | 38 tok/s | 2.85 s | $0.600 | $2.40 | 205k | fp8 |
| GMICloud | 35 tok/s | 1.16 s | $0.300 | $1.20 | 197k | fp8 |
| Venice | 19 tok/s | 1.19 s | $0.270 | $0.950 | 198k | – |
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 $0.030 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0005 | $0.0004 | 36.8 s |
| Summarize a 30-page report | 12,000 / 600 | $0.0043 | $0.0019 | 41.4 s |
| Code edit | 6,000 / 1,500 | $0.0036 | $0.0024 | 55.2 s |
| Agentic coding session | 60,000 / 4,000 | $0.023 | $0.011 | 1.6 min |
| Structured extraction | 2,000 / 200 | $0.0008 | $0.0004 | 35.3 s |
See also
Questions people ask
- What are MiniMax M2.5's benchmark scores?
- As of 25 Oct 2026, MiniMax M2.5 scores 53.6 ±4.3 on the BenchLeader Index, strongest in instruction following (70) and weakest in reasoning (45). That rests on 24 independent benchmark results, and every score on this page links to its publisher.
- How does MiniMax M2.5 rank?
- MiniMax M2.5 is #280 of 785 ranked configurations on 25 Oct 2026. 297 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 M2.5 cost per 1M tokens?
- As of 25 Oct 2026, MiniMax M2.5 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 M2.5 released?
- MiniMax released MiniMax M2.5 on 12 Feb 2026. BenchLeader has results for it at 2 reasoning-effort settings.
Data as of 25 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “MiniMax M2.5: benchmarks, pricing, speed and rank”, data as of 25 Oct 2026, https://www.benchleader.com/models/minimax-m2-5.