MiniMax, open weights
MiniMax-M2.5
MiniMax-M2.5 by MiniMax ranks 188th of 354 ranked models on the Noometry Index as of October 2026, with a score of 38.3. Its strongest category is coding, where it ranks 58th. API pricing starts at $0.30 per million input tokens and $1.20 per million output tokens, with a 205K-token context window.
Last verified
Specifications
- Noometry rank
- #188 of 354
- Index score
- 38.3
- Evidence
- Confirmed 33 results
- Provider
MiniMax- Released
- February 12, 2026
- Weights
- Open weights
- Reasoning
- Yes
- Context window
- 205K
- Max output
- 131K
- Input price
- $0.30 / M
- Output price
- $1.20 / M
- Blended price
- $0.52 / M
- Output speed
- 49 tokens/s Kagi
- Value
- #80 of 219
- Knowledge cutoff
- June 2025
- Input
- text
- Hugging Face
- MiniMaxAI/MiniMax-M2.5
Category scores
Each category score combines every public result we have in that category.
- Coding 48.1
- Agentic & Tool Use 30.4
- Reasoning 17.5
- Math 26.9
- Knowledge 39.2
- Multilingual 47.1
- Instruction Following 71.5
- Long Context 37.5
- Writing & Preference 53.9
| Category | Score | Rank | Results |
|---|---|---|---|
| Coding | 48.1 | #58 | 4 |
| Agentic & Tool Use | 30.4 | #77 | 1 |
| Reasoning | 17.5 | #292 | 5 |
| Math | 26.9 | #253 | 2 |
| Knowledge | 39.2 | #135 | 2 |
| Multilingual | 47.1 | #152 | 1 |
| Instruction Following | 71.5 | #148 | 1 |
| Long Context | 37.5 | #216 | 3 |
| Writing & Preference | 53.9 | #153 | 4 |
Strengths and weaknesses
Categories where MiniMax-M2.5 places highest and lowest among the models ranked in each, with its score against that category's median.
Strongest categories
| Category | Score | vs median | Rank |
|---|---|---|---|
| Coding | 48.1 | +9.3 | #58 of 340, top 18% |
| Knowledge | 39.2 | +1.9 | #135 of 314, top 43% |
| Instruction Following | 71.5 | +0.2 | #148 of 305, top 49% |
Weakest categories
| Category | Score | vs median | Rank |
|---|---|---|---|
| Reasoning | 17.5 | −6.1 | #292 of 350, top 84% |
| Math | 26.9 | −9.6 | #253 of 327, top 78% |
| Long Context | 37.5 | −3.5 | #216 of 296, top 73% |
Closest competitors
The models ranked just above and below MiniMax-M2.5. When scores are this close, price and speed are often the better way to choose.
| Model | Rank | Score | Blended $/M | Speed | |
|---|---|---|---|---|---|
| Hunyuan Large 2025 02 10 | #184 | 38.6 | — | — | Compare |
| Trinity Large Thinking | #185 | 38.6 | $0.39 | — | Compare |
| GPT-5.1-Codex | #186 | 38.6 | $3.44 | — | Compare |
| Sonar | #187 | 38.5 | $1 | — | Compare |
| Nova Premier 1.0 | #189 | 38.3 | $5 | 9 | Compare |
| Qwen3-Coder 480B-A35B Instruct | #190 | 38.1 | $3 | 67 | Compare |
| Olmo 3.1 32b Think | #191 | 37.9 | — | — | Compare |
| GPT-5-Codex | #192 | 37.9 | $3.44 | 95 | Compare |
Sponsored placements are available on pages like this one. Advertise on Noometry
Benchmark results
Every published result we track, with its source. Bold rows are the ones used for ranking; where several exist we prefer independent runs over self-reported numbers.
Coding
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| SWE-bench Verified (bash only) | 75.8% | #3 of 39, top 8% | high | SWE-bench | 2026-02-17 |
| LMArena WebDev | 1387 | #78 of 113, top 70% | LMArena | 2026-10-08 | |
| SWE-bench Multilingual | 68.3% | #6 of 13, top 47% | SWE-bench | 2026-02-16 | |
| LMArena Coding | 1381 | #148 of 294, top 51% | LMArena | 2026-10-08 | |
| ALE-Bench | 618.17 | #79 of 105, top 76% | Epoch AI |
Agentic & Tool Use
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| Terminal-Bench | 42.7% | #23 of 41, top 57% | Epoch AI | ||
| Vending-Bench 2 | -23.16 | #59 of 60, top 99% | Epoch AI |
Reasoning
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| ARC-AGI-2 | 4.9% | #58 of 83, top 70% | Epoch AI | ||
| Kagi LLM Benchmark | 55.2% | #51 of 99, top 52% | Kagi LLM Benchmark | ||
| NYT Connections (extended) | 16.8% | #80 of 91, top 88% | Lech Mazur benchmarks | ||
| ARC-AGI-1 | 63.7% | #49 of 83, top 60% | Epoch AI | ||
| LMArena Hard Prompts | 1372 | #145 of 297, top 49% | LMArena | 2026-10-08 | |
| Epoch Capabilities Index | 146.68 | #70 of 213, top 33% | Epoch AI | 2026-02-12 |
Math
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| ProofBench | 4% | #71 of 77, top 93% | Epoch AI | ||
| LMArena Math | 1378 | #145 of 285, top 51% | LMArena | 2026-10-08 |
Knowledge
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| Vectara Hallucination Rate (lower is better) | 9.1% | #42 of 96, top 44% | Vectara Hallucination Leaderboard | ||
| LMArena Expert | 1379 | #138 of 273, top 51% | LMArena | 2026-10-08 |
Multilingual
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| LMArena Non-English | 1338 | #152 of 297, top 52% | LMArena | 2026-10-08 | |
| LMArena Chinese | 1393 | #141 of 285, top 50% | LMArena | 2026-10-08 | |
| LMArena French | 1362 | #141 of 223, top 64% | LMArena | 2026-10-08 | |
| LMArena German | 1362 | #123 of 231, top 54% | LMArena | 2026-10-08 | |
| LMArena Japanese | 1171 | #169 of 211, top 81% | LMArena | 2026-10-08 | |
| LMArena Korean | 1232 | #151 of 213, top 71% | LMArena | 2026-10-08 | |
| LMArena Russian | 1358 | #142 of 283, top 51% | LMArena | 2026-10-08 | |
| LMArena Spanish | 1354 | #141 of 226, top 63% | LMArena | 2026-10-08 |
Instruction Following
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| LMArena Instruction Following | 1353 | #143 of 298, top 48% | LMArena | 2026-10-08 |
Long Context
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| CL-bench | 11.4% | #19 of 19, top 100% | Epoch AI | ||
| CL-bench Life | 6.3% | #13 of 13, top 100% | Epoch AI | ||
| LMArena Longer Query | 1366 | #141 of 291, top 49% | LMArena | 2026-10-08 |
Writing & Preference
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| LMArena Text | 1359 | #151 of 297, top 51% | LMArena | 2026-10-08 | |
| LMArena Creative Writing | 1331 | #148 of 295, top 51% | LMArena | 2026-10-08 | |
| EQ-Bench Creative Writing | 1361 | #74 of 115, top 65% | EQ-Bench | ||
| LMArena Multi-Turn | 1364 | #146 of 295, top 50% | LMArena | 2026-10-08 |
API pricing by provider
| Route | Input $/M | Output $/M | Cached input $/M | Checked |
|---|---|---|---|---|
| bedrock | $0.30 | $1.20 | — | 2026-10-10 |
| deepinfra | $0.15 | $1.15 | $0.03 | 2026-10-10 |
| minimax | $0.30 | $1.20 | $0.03 | 2026-10-10 |
| openrouter | $0.27 | $1.08 | $0.027 | 2026-10-10 |
| together | $0.30 | $1.20 | $0.06 | 2026-10-10 |
Compare MiniMax-M2.5
- MiniMax-M2.5 vs MiniMax-M2.1
- MiniMax-M2.5 vs Sonar
- MiniMax-M2.5 vs Nova Premier 1.0
- MiniMax-M2.5 vs GPT-5.1-Codex
- MiniMax-M2.5 vs Qwen3-Coder 480B-A35B Instruct
- MiniMax-M2.5 vs Trinity Large Thinking
- MiniMax-M2.5 vs Olmo 3.1 32b Think
- MiniMax-M2.5 vs GPT-6 Astra
- MiniMax-M2.5 vs Claude Fable 5.1
- MiniMax-M2.5 vs Gemini 3.8 Flash
- MiniMax-M2.5 vs Kimi K3
- MiniMax-M2.5 vs Grok 4.6
- MiniMax-M2.5 vs Qwen3.8 Max
- MiniMax-M2.5 vs GLM-5.3
Other MiniMax models
Frequently asked questions
How good is MiniMax-M2.5?
MiniMax-M2.5 by MiniMax ranks 188th of 354 ranked models on the Noometry Index as of October 2026, with a score of 38.3. Its strongest category is coding, where it ranks 58th. API pricing starts at $0.30 per million input tokens and $1.20 per million output tokens, with a 205K-token context window.
How much does MiniMax-M2.5 cost?
MiniMax-M2.5 costs $0.30 per million input tokens and $1.20 per million output tokens on MiniMax's own API, with cached input at $0.03.
What is MiniMax-M2.5's context window?
MiniMax-M2.5 accepts up to 205K tokens of input and can write up to 131K tokens in one response.
Is MiniMax-M2.5 open source?
Yes. MiniMax-M2.5's weights are downloadable from Hugging Face (MiniMaxAI/MiniMax-M2.5); check the license for commercial terms.
How fast is MiniMax-M2.5?
MiniMax-M2.5 generated about 49 output tokens per second in the Kagi LLM Benchmark's timed runs. Speed varies by provider, load and reasoning effort.
What are MiniMax-M2.5's strengths and weaknesses?
Relative to other ranked models, MiniMax-M2.5 places best in coding, knowledge, instruction following and lowest in reasoning, math, long context.
What is MiniMax-M2.5 best at?
Its best category is coding, where it ranks 58th on Noometry.