MiniMax-M2.7vsMiniMax M2.5
Across 9 shared benchmarks, MiniMax-M2.7 leads overall: MiniMax-M2.7 wins 5, MiniMax M2.5 wins 4, with 0 ties and an average score difference of +1.87.
MiniMax-M2.7
MiniMaxAI · 2026-03-18 · Reasoning model
MiniMax M2.5
MiniMaxAI · 2026-02-12 · Reasoning model
MiniMax-M2.75 wins(56%)(44%)4 winsMiniMax M2.5
Benchmark scores
Grouped by capability, sorted by largest gap within each. 9 shared benchmarks.
Claw-style Agent Evaluation
MiniMax M2.5 2/2| Benchmark | MiniMax-M2.7 | MiniMax M2.5 | Diff |
|---|---|---|---|
| Pinch Bench | 87.109 / 37Thinking (With Tools) | 87.806 / 37Thinking (With Tools) | -0.70 |
| Claw Bench | 91.705 / 29Thinking (With Tools) | 92.104 / 29Thinking (With Tools) | -0.40 |
General Knowledge
MiniMax-M2.7 2/2| Benchmark | MiniMax-M2.7 | MiniMax M2.5 | Diff |
|---|---|---|---|
| HLE | 2882 / 157Thinking (No Tools) | 19.40106 / 157Thinking (No Tools) | +8.60 |
| GPQA Diamond | 8738 / 178Thinking (No Tools) | 85.2048 / 178Thinking (No Tools) | +1.80 |
Agent Level Benchmark
MiniMax M2.5 1/1| Benchmark | MiniMax-M2.7 | MiniMax M2.5 | Diff |
|---|---|---|---|
| τ²-Bench - Telecom | 8524 / 35Thinking (With Tools) | 97.8010 / 35 | -12.80 |
Coding and Software Engineer
MiniMax-M2.7 1/1| Benchmark | MiniMax-M2.7 | MiniMax M2.5 | Diff |
|---|---|---|---|
| SWE-Bench Pro - Public | 56.2016 / 43Thinking (With Tools) | 55.4018 / 43 | +0.80 |
Instruction Following
MiniMax-M2.7 1/1| Benchmark | MiniMax-M2.7 | MiniMax M2.5 | Diff |
|---|---|---|---|
| IF Bench | 765 / 29Thinking (With Tools) | 7012 / 29 | +6 |
Long Context
MiniMax M2.5 1/1| Benchmark | MiniMax-M2.7 | MiniMax M2.5 | Diff |
|---|---|---|---|
| AA-LCR | 694 / 13Thinking (With Tools) | 69.503 / 13Thinking (No Tools) | -0.50 |
Productivity Knowledge
MiniMax-M2.7 1/1| Benchmark | MiniMax-M2.7 | MiniMax M2.5 | Diff |
|---|---|---|---|
| GDPval-AA | 5013 / 21Thinking (No Tools) | 3617 / 21Thinking (No Tools) | +14 |
Specs
| Field | MiniMax-M2.7 | MiniMax M2.5 |
|---|---|---|
| Publisher | MiniMaxAI | MiniMaxAI |
| Release date | 2026-03-18 | 2026-02-12 |
| Model type | Reasoning model | Reasoning model |
| Architecture | MoE | MoE |
| Parameters | 229B | 229B |
| Context length | 200K | 128K |
| Max output | 200K | Not available |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | MiniMax-M2.7 | MiniMax M2.5 |
|---|---|---|
| Text input | $0.3 / 1M tokens | $0.3 / 1M tokens |
| Text output | $1.2 / 1M tokens | $2.4 / 1M tokens |
| Cache read | $0.06 / 1M tokens | Not public |
| Cache write | $0.375 / 1M tokens | Not public |
Summary
- MiniMax-M2.7leads in:General Knowledge (2/2), Coding and Software Engineer (1/1), Instruction Following (1/1), Productivity Knowledge (1/1)
- MiniMax M2.5leads in:Claw-style Agent Evaluation (2/2), Agent Level Benchmark (1/1), Long Context (1/1)
On average across the 9 shared benchmarks, MiniMax-M2.7 scores 1.87 higher.
Largest single-benchmark gap: GDPval-AA — MiniMax-M2.7 50 vs MiniMax M2.5 36 (+14).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.