MiniMax-M2.7vsM2.1

Across 6 shared benchmarks, MiniMax-M2.7 leads overall: MiniMax-M2.7 wins 5, M2.1 wins 1, with 0 ties and an average score difference of +7.07.

MiniMaxAI
MiniMax-M2.7

MiniMaxAI · 2026-03-18 · Reasoning model

MiniMaxAI
M2.1

MiniMaxAI · 2025-12-23 · Chat model

MiniMax-M2.75 wins(83%)(17%)1 winM2.1

Benchmark scores

Grouped by capability, sorted by largest gap within each. 6 shared benchmarks.

General Knowledge

MiniMax-M2.7 2/2
BenchmarkMiniMax-M2.7M2.1Diff
GPQA Diamond8738 / 178Thinking (No Tools)8169 / 178+6
HLE2882 / 157Thinking (No Tools)2294 / 157+6

Agent Level Benchmark

M2.1 1/1
BenchmarkMiniMax-M2.7M2.1Diff
τ²-Bench - Telecom8524 / 35Thinking (With Tools)8722 / 35-2

Claw-style Agent Evaluation

MiniMax-M2.7 1/1
BenchmarkMiniMax-M2.7M2.1Diff
Pinch Bench87.109 / 37Thinking (With Tools)84.3018 / 37Thinking (With Tools)+2.80

Coding and Software Engineer

MiniMax-M2.7 1/1
BenchmarkMiniMax-M2.7M2.1Diff
SWE-Bench Pro - Public56.2016 / 43Thinking (With Tools)32.6042 / 43+23.60

Instruction Following

MiniMax-M2.7 1/1
BenchmarkMiniMax-M2.7M2.1Diff
IF Bench765 / 29Thinking (With Tools)7012 / 29+6

Specs

FieldMiniMax-M2.7M2.1
PublisherMiniMaxAIMiniMaxAI
Release date2026-03-182025-12-23
Model typeReasoning modelChat model
ArchitectureMoEMoE
Parameters229B230B
Context length200K200K
Max output200K128K

API pricing

Prices use DataLearner records when available; missing fields are not inferred.

ItemMiniMax-M2.7M2.1
Text input$0.3 / 1M tokensNot public
Text output$1.2 / 1M tokensNot public
Cache read$0.06 / 1M tokensNot public
Cache write$0.375 / 1M tokensNot public

One or both models have incomplete public pricing.

Summary

  • MiniMax-M2.7leads in:General Knowledge (2/2), Claw-style Agent Evaluation (1/1), Coding and Software Engineer (1/1), Instruction Following (1/1)
  • M2.1leads in:Agent Level Benchmark (1/1)

On average across the 6 shared benchmarks, MiniMax-M2.7 scores 7.07 higher.

Largest single-benchmark gap: SWE-Bench Pro - Public — MiniMax-M2.7 56.20 vs M2.1 32.60 (+23.60).

Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.