Model comparison
MiniMax-M2.5 vs MiniMax-M3
MiniMax-M3 is the stronger model overall, scoring 43.5 to 38.1 on the Noometry Index.
Last verified . 23 shared benchmarks.
MiniMaxSummary
- They share 23 benchmarks with published results for both. MiniMax-M2.5 scores higher in 2 categories and MiniMax-M3 in 7 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in knowledge, where MiniMax-M3 leads 58.4 to 38.6.
- The biggest single-benchmark swing is NYT Connections (extended): 16.8% for MiniMax-M2.5 and 65.1% for MiniMax-M3.
- Both cost about the same: $0.30 input and $1.20 output per million tokens.
- MiniMax-M3 accepts more context: 1M tokens versus 205K.
Side by side
| MiniMax-M2.5 | MiniMax-M3 | |
|---|---|---|
| Provider | MiniMax | MiniMax |
| Noometry Index | 38.1 | 43.5 |
| Released | 2026-02-12 | 2026-06-01 |
| Weights | Open | Open |
| Context window | 205K | 1M |
| Max output | 131K | 512K |
| Input $ / M tokens | $0.30 | $0.30 |
| Output $ / M tokens | $1.20 | $1.20 |
| Results tracked | 33 | 41 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding MiniMax-M2.5 leads
MiniMax-M2.5: 47.9 (#53), MiniMax-M3: 41.5 (#111)
| Benchmark | MiniMax-M2.5 | MiniMax-M3 |
|---|---|---|
| LMArena WebDev | 1386 | 1482 |
| LMArena Coding | 1380 | 1470 |
| ALE-Bench | 618.17 | 640.02 |
| FrontierCode | — | 14.7% |
| SWE-bench Verified (bash only) | 75.8% | — |
| SWE-bench Multilingual | 68.3% | — |
| SciCode | — | 47.1% |
Agentic & Tool Use MiniMax-M2.5 leads
MiniMax-M2.5: 30.2 (#70), MiniMax-M3: 22.5 (#121)
| Benchmark | MiniMax-M2.5 | MiniMax-M3 |
|---|---|---|
| Vending-Bench 2 | -23.16 | 2,158 |
| Terminal-Bench | 42.7% | — |
| APEX-Agents | — | 37.7% |
| OSWorld 2.0 | — | 4.6% |
| GBAEval | — | 0.9% |
Reasoning MiniMax-M3 leads
MiniMax-M2.5: 17.6 (#281), MiniMax-M3: 29.3 (#83)
| Benchmark | MiniMax-M2.5 | MiniMax-M3 |
|---|---|---|
| NYT Connections (extended) | 16.8% | 65.1% |
| LMArena Hard Prompts | 1372 | 1447 |
| Epoch Capabilities Index | 146.68 | 147 |
| ARC-AGI-2 | 4.9% | — |
| SimpleBench | — | 45.8% |
| Kagi LLM Benchmark | 55.2% | — |
| ARC-AGI-1 | 63.7% | — |
| CritPt | — | 3.7% |
| Chess Puzzles | — | 14% |
| Mystery Game Puzzles | — | 8% |
| DTBench | — | 78.9% |
| LMCA | — | 33.7% |
| Surface Evolver Bench | — | 55% |
| ForecastBench | — | 61.4 |
Math MiniMax-M3 leads
MiniMax-M2.5: 25.8 (#242), MiniMax-M3: 39.1 (#83)
| Benchmark | MiniMax-M2.5 | MiniMax-M3 |
|---|---|---|
| ProofBench | 4% | 18% |
| LMArena Math | 1378 | 1432 |
| OTIS Mock AIME 2024-2025 | — | 71.1% |
Knowledge MiniMax-M3 leads
MiniMax-M2.5: 38.6 (#132), MiniMax-M3: 58.4 (#32)
| Benchmark | MiniMax-M2.5 | MiniMax-M3 |
|---|---|---|
| LMArena Expert | 1378 | 1462 |
| GPQA Diamond | — | 90.9% |
| Vectara Hallucination Rate | 9.1% | — |
Multimodal Not comparable
MiniMax-M2.5: —, MiniMax-M3: 39.5 (#45)
| Benchmark | MiniMax-M2.5 | MiniMax-M3 |
|---|---|---|
| LMArena Vision | — | 1254 |
| LMArena Document | — | 1435 |
Multilingual MiniMax-M3 leads
MiniMax-M2.5: 47.3 (#147), MiniMax-M3: 53.2 (#70)
| Benchmark | MiniMax-M2.5 | MiniMax-M3 |
|---|---|---|
| LMArena Non-English | 1338 | 1420 |
| LMArena Chinese | 1392 | 1465 |
| LMArena French | 1363 | 1445 |
| LMArena German | 1361 | 1424 |
| LMArena Japanese | 1169 | 1381 |
| LMArena Korean | 1232 | 1372 |
| LMArena Russian | 1359 | 1430 |
| LMArena Spanish | 1354 | 1431 |
Instruction Following MiniMax-M3 leads
MiniMax-M2.5: 71.7 (#143), MiniMax-M3: 75.7 (#56)
| Benchmark | MiniMax-M2.5 | MiniMax-M3 |
|---|---|---|
| LMArena Instruction Following | 1353 | 1433 |
Long Context MiniMax-M3 leads
MiniMax-M2.5: 37.4 (#211), MiniMax-M3: 44.2 (#68)
| Benchmark | MiniMax-M2.5 | MiniMax-M3 |
|---|---|---|
| LMArena Longer Query | 1367 | 1445 |
| CL-bench | 11.4% | — |
| CL-bench Life | 6.3% | — |
Writing & Preference MiniMax-M3 leads
MiniMax-M2.5: 54.1 (#149), MiniMax-M3: 62.3 (#79)
| Benchmark | MiniMax-M2.5 | MiniMax-M3 |
|---|---|---|
| LMArena Text | 1359 | 1433 |
| LMArena Creative Writing | 1331 | 1405 |
| LMArena Multi-Turn | 1364 | 1441 |
| EQ-Bench Creative Writing | 1361 | — |
| EQ-Bench 4 | — | 1150 |
Frequently asked questions
Is MiniMax-M2.5 better than MiniMax-M3?
MiniMax-M3 is the stronger model overall, scoring 43.5 to 38.1 on the Noometry Index.
Which is cheaper, MiniMax-M2.5 or MiniMax-M3?
MiniMax-M3 is cheaper. It lists at $0.30 per million input tokens and $1.20 per million output tokens; MiniMax-M2.5 lists at $0.30 and $1.20.
Is MiniMax-M2.5 or MiniMax-M3 better for coding?
MiniMax-M2.5 scores higher on coding benchmarks: 47.9 versus 41.5 in the Noometry coding category.
Which has the bigger context window?
MiniMax-M3 does, with 1M tokens against 205K.
How many benchmarks do MiniMax-M2.5 and MiniMax-M3 share?
23 benchmarks have published results for both models. MiniMax-M2.5 has 33 scored results on Noometry and MiniMax-M3 has 41.