Benchmark Almanac

MiniMax-M2

MiniMax-M2 ranks first on ArtifactsBench.

MiniMax Released October 27th, 2025 Family MiniMax-M2 The lab's model card Compare with others
44
Benchmarks with a score
1
Ranked first, out of five or more models
1
In the top three
MiniMax Oct 2025

MiniMax-M2

Open weights Flagship

141.0 #169 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
  • IFBench 72% #33 /80
  • Terminal-Bench 2.0 30% #32 /43
  • SciCode 36% #39 /49

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

BenchmarkScoreRankSource
MMLU-Pro 82% #59 of 167 Lab model card
IFBench 72% #33 of 80 Lab model card
Humanity's Last Exam 31.8% #75 of 154 Lab model card
LMArena (Elo) 1,344 #183 of 335 Official leaderboard
AA-LCR (Artificial Analysis Long Context Reasoning) 61% #23 of 38 Lab model card
Longform Creative Writing (EQ-Bench) 46.4 #90 of 119 Official leaderboard
WideSearch 62.3% #15 of 19 Lab model card
Extended NYT Connections 14.8% #93 of 103 Official leaderboard

Math and science

BenchmarkScoreRankSource
GPQA Diamond 78% #131 of 302 Lab model card
AIME 2025 78% #97 of 143 Lab model card
SimpleBench 25% #92 of 106 Official leaderboard

Coding

BenchmarkScoreRankSource
LiveCodeBench 83% #9 of 54 Lab model card
ArtifactsBench 66.8% #1 of 5 Lab model card
SWE-bench Verified 61% #28 of 60 Official leaderboard
MLE-bench 40% #5 of 10 Lab model card
Multi-SWE-bench 36.2% #7 of 10 Lab model card
SciCode 36% #39 of 49 Lab model card
CooperBench 14% #5 of 6 Official leaderboard
Code Arena 1,299 #110 of 128 Official leaderboard
SWE-bench Multilingual 56.5% #63 of 73 Lab model card
WebDev Arena 1,298 #91 of 105 arena.ai, via Epoch AI
OctoCodingbench 13.3% #6 of 6 Official leaderboard
SWE-Review 3.40% #5 of 5 Official leaderboard
VIBE (MiniMax) 67.5% #5 of 5 Official leaderboard
PRDBench 24.8% #4 of 4 Official leaderboard
VIBE-Pro (MiniMax) 42.4% #3 of 3 Lab model card

Agents and tools

BenchmarkScoreRankSource
tau2-bench 87% #21 of 79 Lab model card
Elimination Game 5 #19 of 56 Official leaderboard
Terminal-Bench 30% #8 of 18 tbench.ai, via Epoch AI
SnakeBench 24 #93 of 188 Official leaderboard
Terminal-Bench Hard 24% #8 of 13 Lab model card
Terminal-Bench 2.0 30% #32 of 43 Official leaderboard
GAIA 75.7% #4 of 5 Lab model card
BrowseComp 44% #69 of 81 Lab model card
BrowseComp-ZH 48.5% #17 of 19 Lab model card
Toolathlon 18.8% #29 of 29 Lab model card
xbench-DeepSearch 72% #3 of 3 Lab model card

Professional work

BenchmarkScoreRankSource
Physical-Engineering's Last Exam 0.06 #14 of 18 Official leaderboard
Vending-Bench 2 $161 #60 of 67 Official leaderboard
ProfBench 37.7% #66 of 72 Official leaderboard
FinSearchComp Global 65.5% #1 of 1 Lab model card
TheAgentCompany 36% #1 of 1 Lab model card

Safety and security

BenchmarkScoreRankSource
LiveSecBench 66.7 #8 of 34 Official leaderboard

Other

BenchmarkScoreRankSource
Artificial Analysis Intelligence Index v3 61.0 #1 of 1 Lab model card

Its lab's numbers put it at 141.0, number 169 of 404

The Self-Reported Index fits one score to the 18 capability results MiniMax reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 139.0 to 143.0. How we compute it.

Its lab's numbers next to 2 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark Lab says Board says Gap
SWE-bench Verified 69.4% SWE-bench Verified 61% Leaderboard, mini-SWE-agent +8.4 pts
MMLU-Pro 82% MMLU-Pro (AA) 82% Leaderboard Same

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

MiniMax-M2 model card (Hugging Face) Oct 27, 2025

Its system card

Every result in the longer documents the lab published with it. 5 of them aren't in the launch post.