Model comparison
DeepSeek-V3 vs MiniMax-M2.5
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 38.3 on the Noometry Index.
Last verified . 21 shared benchmarks.
DeepSeek
MiniMaxSummary
- They share 21 benchmarks with published results for both. DeepSeek-V3 scores higher in 5 categories and MiniMax-M2.5 in 3 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in coding, where MiniMax-M2.5 leads 48.1 to 42.3.
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $0.30 / $1.20 for MiniMax-M2.5.
- MiniMax-M2.5 accepts more context: 205K tokens versus 164K.
Side by side
| DeepSeek-V3 | MiniMax-M2.5 | |
|---|---|---|
| Provider | DeepSeek | MiniMax |
| Noometry Index | 39.5 | 38.3 |
| Released | 2024-12-26 | 2026-02-12 |
| Weights | Open | Open |
| Context window | 164K | 205K |
| Max output | 164K | 131K |
| Input $ / M tokens | $0.24 | $0.30 |
| Output $ / M tokens | $0.90 | $1.20 |
| Results tracked | 60 | 33 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding MiniMax-M2.5 leads
DeepSeek-V3: 42.3 (#106), MiniMax-M2.5: 48.1 (#58)
| Benchmark | DeepSeek-V3 | MiniMax-M2.5 |
|---|---|---|
| LMArena Coding | 1368 | 1381 |
| SWE-bench Verified (bash only) | — | 75.8% |
| Aider Polyglot | 55.1% | — |
| LMArena WebDev | — | 1387 |
| SWE-bench Multilingual | — | 68.3% |
| SciCode | 35.8% | — |
| WeirdML | 36.1% | — |
| BigCodeBench Instruct | 50% | — |
| LiveBench Coding | 70.9% | — |
| BigCodeBench Complete | 62.2% | — |
| ALE-Bench | — | 618.17 |
| HumanEval+ | 86.6% | — |
| MBPP+ | 73% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, MiniMax-M2.5: 30.4 (#77)
| Benchmark | DeepSeek-V3 | MiniMax-M2.5 |
|---|---|---|
| Terminal-Bench | — | 42.7% |
| METR Time Horizons | 49.6% | — |
| Vending-Bench 2 | — | -23.16 |
Reasoning DeepSeek-V3 leads
DeepSeek-V3: 20.5 (#236), MiniMax-M2.5: 17.5 (#292)
| Benchmark | DeepSeek-V3 | MiniMax-M2.5 |
|---|---|---|
| Kagi LLM Benchmark | 52.3% | 55.2% |
| LMArena Hard Prompts | 1365 | 1372 |
| Epoch Capabilities Index | 135.94 | 146.68 |
| ARC-AGI-2 | — | 4.9% |
| SimpleBench | 27.2% | — |
| NYT Connections (extended) | — | 16.8% |
| ARC-AGI-1 | — | 63.7% |
| CritPt | 0% | — |
| LiveBench Reasoning | 65.8% | — |
| DTBench | 64.8% | — |
| LiveBench Data Analysis | 60.9% | — |
| LMCA | 15.5% | — |
| BIG-Bench Hard | 87.5% | — |
| ForecastBench | 59.1 | — |
| HellaSwag | 88.9% | — |
| LiveBench | 66.9% | — |
| PIQA | 84.7% | — |
| WinoGrande | 85.2% | — |
Math DeepSeek-V3 leads
DeepSeek-V3: 32.1 (#219), MiniMax-M2.5: 26.9 (#253)
| Benchmark | DeepSeek-V3 | MiniMax-M2.5 |
|---|---|---|
| LMArena Math | 1373 | 1378 |
| OTIS Mock AIME 2024-2025 | 37.8% | — |
| ProofBench | — | 4% |
| Omni-MATH | 40.3% | — |
| LiveBench Math | 73.5% | — |
| MATH Level 5 | 75.5% | — |
| FrontierMath (Feb 2025 set) | 1.7% | — |
Knowledge MiniMax-M2.5 leads
DeepSeek-V3: 37.5 (#155), MiniMax-M2.5: 39.2 (#135)
| Benchmark | DeepSeek-V3 | MiniMax-M2.5 |
|---|---|---|
| Vectara Hallucination Rate | 6.1% | 9.1% |
| LMArena Expert | 1351 | 1379 |
| GPQA Diamond | 67.6% | — |
| MMLU-Pro | 72.3% | — |
| Confabulations | 26.1% | — |
| GPQA (HELM) | 53.8% | — |
| ARC (AI2) Challenge | 95.3% | — |
| MMLU | 87.2% | — |
| TriviaQA | 82.9% | — |
Multilingual DeepSeek-V3 leads
DeepSeek-V3: 48.5 (#143), MiniMax-M2.5: 47.1 (#152)
| Benchmark | DeepSeek-V3 | MiniMax-M2.5 |
|---|---|---|
| LMArena Non-English | 1358 | 1338 |
| LMArena Chinese | 1391 | 1393 |
| LMArena French | 1385 | 1362 |
| LMArena German | 1374 | 1362 |
| LMArena Japanese | 1333 | 1171 |
| LMArena Korean | 1319 | 1232 |
| LMArena Russian | 1373 | 1358 |
| LMArena Spanish | 1358 | 1354 |
Instruction Following DeepSeek-V3 leads
DeepSeek-V3: 72.8 (#130), MiniMax-M2.5: 71.5 (#148)
| Benchmark | DeepSeek-V3 | MiniMax-M2.5 |
|---|---|---|
| LMArena Instruction Following | 1345 | 1353 |
| LiveBench Instruction Following | 81.5% | — |
| IFEval | 83.2% | — |
Long Context MiniMax-M2.5 leads
DeepSeek-V3: 34.0 (#253), MiniMax-M2.5: 37.5 (#216)
| Benchmark | DeepSeek-V3 | MiniMax-M2.5 |
|---|---|---|
| LMArena Longer Query | 1352 | 1366 |
| Fiction.LiveBench | 50% | — |
| CL-bench | — | 11.4% |
| CL-bench Life | — | 6.3% |
Writing & Preference DeepSeek-V3 leads
DeepSeek-V3: 57.4 (#130), MiniMax-M2.5: 53.9 (#153)
| Benchmark | DeepSeek-V3 | MiniMax-M2.5 |
|---|---|---|
| LMArena Text | 1375 | 1359 |
| LMArena Creative Writing | 1364 | 1331 |
| EQ-Bench Creative Writing | 1472 | 1361 |
| LMArena Multi-Turn | 1389 | 1364 |
| Short-Story Creative Writing | 77% | — |
| WildBench | 83% | — |
| LiveBench Language | 49.1% | — |
Frequently asked questions
Is DeepSeek-V3 better than MiniMax-M2.5?
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 38.3 on the Noometry Index.
Which is cheaper, DeepSeek-V3 or MiniMax-M2.5?
DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; MiniMax-M2.5 lists at $0.30 and $1.20.
Is DeepSeek-V3 or MiniMax-M2.5 better for coding?
MiniMax-M2.5 scores higher on coding benchmarks: 48.1 versus 42.3 in the Noometry coding category.
Which has the bigger context window?
MiniMax-M2.5 does, with 205K tokens against 164K.
How many benchmarks do DeepSeek-V3 and MiniMax-M2.5 share?
21 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and MiniMax-M2.5 has 33.