MiniMax M3
MiniMax M3MiniMaxOpen weightsNon-reasoningVerifiedReleased Jun 1, 2026Input: text, image, video
Overall score
55.0
Rank 69 of 222 · 45.6–64.4
Context window
1M
1,000,000 tokens
Price per 1M tokens
$0.30 in · $1.20 out
Source: BenchLM
Blended price
$0.53
3:1 input:output tokens
Output speed
—
Not measured yet
Category scores
Scores per capability area with the model's rank among ranked models in that category.
- 1Instruction following#593.8
- 2Reasoning82.0
- 3Multimodal#4460.0
- 4Agentic & tool use#7059.0
- 5Coding#7457.3
- 6Knowledge#7652.3
Benchmark results (24)
As reported in system cards, papers and public evaluations. Position compares this model with every other model reporting the same benchmark.
Agentic & tool use
| Terminal-Bench 2.1 | 66.0 | #46 of 68 |
| BrowseComp | 83.5 | #17 of 50 |
| MCP Atlas | 74.2 | #19 of 42 |
| MazeBench | 0 | #20 of 42 |
| Claw-Eval | 74.5 | #4 of 36 |
| OSWorld-Verified | 70.1 | #23 of 35 |
| OSWorld 2.0 | 4.6 | #22 of 24 |
| ResearchClawBench | 19.8 | #5 of 19 |
| BankerToolBench | 76.1 | #3 of 3 |
| GDPval rubrics | 74.7 | #1 of 1 |
Coding
| SWE-bench Verified | 80.5 | #15 of 98 |
| SWE-bench Pro | 59.0 | #30 of 69 |
| NL2Repo | 42.1 | #22 of 31 |
| OpenHarmony Bench | 48.4 | #12 of 12 |
| VIBE V2 | 50.1 | #1 of 1 |
| SVG-Bench | 63.7 | #1 of 1 |
| KernelBench Hard | 28.8 | #1 of 1 |
Multimodal
| Design Arena Website | 1258 | #37 of 91 |
| MMMU-Pro | 78.1 | #21 of 48 |
| OfficeQA Pro | 45.1 | #12 of 13 |
| VideoMMMU | 84.6 | #8 of 12 |
| OmniDocBench 1.5 | 91.6 | #2 of 11 |
| Video-MME (with subtitle) | 85.4 | #10 of 11 |
Math
| USAMO 2026 | 85.7 | #3 of 3 |
Pricing by source
USD per 1M tokens.
| Source | Input | Output | Cached in |
|---|---|---|---|
| List price (BenchLM) | $0.30 | $1.20 | $0.060 |
| OpenRouter minimax/minimax-m3 | $0.30 | $1.20 | $0.060 |
MiniMax's official API pricing page lists the permanent 50%-off standard rate for MiniMax-M3 at $0.30 input / $0.06 cached input / $1.20 output per million tokens for prompts up to 512K. The 512K-to-1M tier is $0.60 / $0.12 / $2.40; BenchLM stores the standard short-context rate.
Arena ratings
Human-preference ratings from LMArena. Best-rated variant shown.
| Text | 1433 | #91 · 64K votes |
| WebDev (coding) | 1482 | #66 · 19.8K votes |
| Vision | 1251 | #62 · 19.8K votes |
| Documents | 1435 | #32 · 6.3K votes |
Open weights
From the Hugging Face repository.
- Parameters
- 427B
- Licence
- other
- Downloads (30 days)
- 200.2K
- Repository
- MiniMaxAI/MiniMax-M3 ↗
Speed
Measured by us through OpenRouter (median of trials).
Not measured yet.
Related models
Same family
- MiniMax M2.748.2
- MiniMax M2.550.0
Similar score
- Qwen 3.6 Max (preview)55.1 · —
- GLM-554.9 · $1.55
- GPT-5.2-Codex55.4 · $4.81
- Qwen3.5 397B55.4 · $1.35
- Agents-A154.4 · —