M
MiniMax M3
Efficient on light hardware with a 1M open-weight context, but essentially no agent ability
Intelligence v5
— v4.3.2 29.2
Leaderboard rank
—
Input / output
$0.25/$1.00
Blended 3:1
$0.44
Output speed
126 tok/s
Value
—
Capability profile Normalised to the field maximum
Specifications
Vendor
MiniMax
Released
2026-06
Weights
Open (MiniMax Community License)
Context
1,000,000 tokens
Max output
Undisclosed
Parameters
428B total / 23B active
Input price
$0.25 / 1M
Output price
$1.00 / 1M
Cache hit
n/a
Output speed
126 tok/s
Intelligence
29.2（v4.3.2）
Benchmark scores
How it is measured →Bars are scaled to the field maximum for that benchmark.
| Benchmarks | Score | Vs. best in field | Gap |
|---|---|---|---|
| AA Intelligence Index v5 | — | — | not indexed |
| AA Intelligence Index v4.3.2 Intelligence index (legacy) | 29.2 | 51% | −28.4 pts |
| GPQA Diamond Graduate-level science reasoning | 90.9 | 94% | −5.5 pts |
| Terminal-Bench 4.0 Terminal-agent tasks | 1.0 | 2% | −64.2 pts |
| LiveBench Design | — | — | not indexed |
| Vals Index Vals composite index | 29.1 | 40% | −44.0 pts |
Assessment
428B/23B with 1M context. Terminal-Bench 4.0 is just 1.0% and APEX-Agents 37.7%, which says it is built for single-turn knowledge questions rather than multi-step tool use. GPQA 90.9 is close to the leaders — a knowledge-first open model.
At the same price
The four models closest on blended price, side by side.
| Model | v5 | Blended | tok/s | Score gap |
|---|---|---|---|---|
| DeepSeek V4.1 Flash Open weightsDeepSeek | 32.5 | $0.53 | 148 | — |
| MiMo-V2.6 Pro Open weightsXiaomi | 30.0 | $0.61 | 81 | — |
| MiniMax M3.1 FlashMiniMax | 29.0 | $0.26 | 118 | — |
| Claude Haiku 5.5Anthropic | 36.2 | $0.20 | 235 | — |
Compare side by side
Line it up against other models on price, speed and every benchmark.
The comparison tool lines up to 4 models side by side with switchable normalisation — built for purchasing decisions.
Open the comparison tool → Back to leaderboard