MiniMaxMiniMax-M2.5
Best configuration ranks #220 of 610 on the BenchLeader Index at 53.5 ±4.4. Last measured 8 Sept 2026. Released 12 Feb 2026.
- Blended price
- $0.525/M
- $0.300 in · $1.20 out
- Output speed
- 93 tok/s
- First answer
- 23 s
- first token 1.04 s
- Context
- 205k
- Overall index54
- Reasoning47
- Coding50
- Agents & tools44
- Maths48
- Knowledge50
- Instruction following70
- Human preference57
- Long context63
- Composite58
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. “Default” means the publisher did not say which setting was used.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| high | – | – | 93 tok/s | 23 s | $0.0005 | Coding 66 |
| defaultbest | 53.5 | #220 | 93 tok/s | 23 s | $0.0005 | Agents & tools 44 · Coding 50 · Composite 58 · Human preference 57 · Instruction following 70 · Knowledge 50 · Long context 63 · Maths 48 · Reasoning 47 |
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 | high | default | Source |
|---|---|---|---|
| LMArena Hard Prompts | – | 1416#138 | LMArena |
| GPQA Diamond (AA)not in index | – | 84.8%#131 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | – | 20.5%#158 | 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%#104 | ARC Prize |
| ARC-AGI-2 | – | 4.9%#120 | ARC Prize |
Coding
| Benchmark | high | default | Source |
|---|---|---|---|
| LMArena Coding | – | 1444#133 | LMArena |
| LMArena WebDev | – | 1384#79 | LMArena |
| LiveCodeBench | – | 79.2%#70 | Vals AI |
| IOI | – | 6.7%#39 | Vals AI |
| SWE-bench (Vals)not in index | – | 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%#5 | – | SWE-bench |
Agents & tools
| Benchmark | high | default | Source |
|---|---|---|---|
| Terminal-Bench | – | 42.7%#35 | Terminal-Bench |
| APEX-Agents | – | 6.2%#56 | Mercor |
| Terminal-Bench Hard | – | 34.9%#80 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | – | 95.3%#24 | Artificial Analysis |
| HiL-Bench | – | 6.3%#16 | Scale AI SEAL |
Maths
| Benchmark | high | default | Source |
|---|---|---|---|
| ProofBench | – | 4.0%#56 | Vals AI |
| AIME (Vals) | – | 88.8%#28 | Vals AI |
Knowledge
| Benchmark | high | default | Source |
|---|---|---|---|
| AA-Omniscience | – | -38.9#251 | Artificial Analysis |
| MMLU-Pro | – | 80.1%#86 | Vals AI |
| LegalBench | – | 80.0%#84 | Vals AI |
| CorpFin | – | 59.6%#74 | Vals AI |
| TaxEval | – | 68.2%#97 | Vals AI |
| MedQA | – | 92.5%#29 | Vals AI |
Instruction following
| Benchmark | high | default | Source |
|---|---|---|---|
| IFBench | – | 71.6%#52 | Artificial Analysis |
Human preference
| Benchmark | high | default | Source |
|---|---|---|---|
| LMArena Text | – | 1391#141 | LMArena |
Long context
| Benchmark | high | default | Source |
|---|---|---|---|
| AA-LCR | – | 73.3%#129 | Artificial Analysis |
Composite
| Benchmark | high | default | Source |
|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | 146.5#59 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | – | 22.8#140 | 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 |
|---|---|---|---|---|---|---|
| Friendli | 128 tok/s | 0.44 s | $0.300 | $1.20 | 197k | – |
| Venice | 68 tok/s | 1.04 s | $0.270 | $0.950 | 198k | – |
| StreamLake | 66 tok/s | 1.04 s | $0.270 | $1.08 | 200k | – |
| DigitalOcean | 64 tok/s | 0.51 s | $0.300 | $1.20 | 66k | – |
| NovitaAI | 55 tok/s | 1.34 s | $0.300 | $1.20 | 205k | fp8 |
| MiniMax Highspeed | 52 tok/s | 1.00 s | $0.600 | $2.40 | 205k | fp8 |
| MiniMax | 50 tok/s | 1.12 s | $0.300 | $1.20 | 205k | fp8 |
| AtlasCloud | 40 tok/s | 2.51 s | $0.295 | $1.20 | 197k | fp8 |
Price history
Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.
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 | 26.5 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0043 | $0.0019 | 29.7 s |
| Code edit | 6,000 / 1,500 | $0.0036 | $0.0024 | 39.4 s |
| Agentic coding session | 60,000 / 4,000 | $0.023 | $0.011 | 1.1 min |
| Structured extraction | 2,000 / 200 | $0.0008 | $0.0004 | 25.4 s |
See also
Data as of 9 Sept 2026. Compare these configurations.