MiniMax model product
MiniMax M3
MiniMax M3 is the first open-weight model to combine three frontier capabilities: top-tier coding and agentic performance, a 1M-token context window, and native multimodality.
Updated Aug 12, 2026. Default version: MiniMax M3
LLMBoard score71.3MiniMax M3
Coverage40%21 benchmark families
Context window512KTokens
Official input price$0.30MiniMax (minimax.io) API
MiniMax M3 Specifications
Technical details for the model's default version.
- Version
- MiniMax M3
- Released
- Jun 1, 2026
- Knowledge cutoff
- Unknown
- Parameters
- 428B
- Context window
- 512K
- Max output
- 128K
- Inputs
- image, text, video
- Outputs
- text
- Open weights
- Yes
- License
- MIT
MiniMax M3 Capability Profile
This profile uses the model's current scored version. Arena ratings and prices are shown separately.
MiniMax M3 Benchmark Results
Benchmark scores for MiniMax M3.
| BankerToolBench | 76.1% | 01 | 1 | 100.0% | C | |
| DRACO | 73.2% | 01 | 1 | 100.0% | C | |
| GDPval-Rubrics | 74.8% | 01 | 1 | 100.0% | C | |
| IMO 2025 | 35 points | 01 | 2 | 100.0% | C | |
| KernelBench Hard | 28.8% | 01 | 1 | 100.0% | C | |
| LiveSQLBench | 40.2% | 01 | 1 | 100.0% | C | |
| LOCA-Bench (256k) | 49.3% | 01 | 1 | 100.0% | C | |
| OmniDocBench 1.5 | 91.6% | 01 | 17 | 100.0% | C | |
| PostTrainBench | 37.1% | 01 | 5 | 100.0% | C | |
| SpreadSheetBench-v1 | 89.3% | 01 | 3 | 100.0% | C | |
| SVG-Bench | 63.7% | 01 | 1 | 100.0% | C | |
| SWE Atlas - Codebase QnA | 37.9% | 01 | 1 | 100.0% | C | |
| SWE Atlas - Test Writing | 30.8% | 01 | 1 | 100.0% | C | |
| SWE-fficiency | 34.8% | 01 | 1 | 100.0% | C | |
| USAMO 2026 | 36 points | 01 | 3 | 100.0% | C | |
| VIBE-V2 | 50.1% | 01 | 1 | 100.0% | C | |
| YC-Bench | 2,100,000 points | 01 | 1 | 100.0% | C | |
| CL-bench | 20.5% | 02 | 2 | 0.0% | C | |
| Claw-Eval | 74.5% | 03 | 13 | 83.3% | C | |
| PaperBench | 52.6% | 03 | 3 | 0.0% | C | |
| APEX-Agents | 27.7% | 06 | 7 | 16.7% | C | |
| Video-MME | 85.4% | 06 | 17 | 68.8% | C | |
| OfficeQA Pro | 45.1% | 07 | 7 | 0.0% | C | |
| VideoMMMU | 84.6% | 08 | 26 | 72.0% | C | |
| Finance Agent v2 | 48.3% | 09 | 26 | 68.0% | B | |
| NL2Repo | 42.1% | 09 | 14 | 38.5% | C | |
| SWE-Bench Verified | 80.5% | 10 | 105 | 91.3% | C | |
| FrontierCode 1.1 | 14.7% | 14 | 15 | 7.1% | B | |
| MCP Atlas | 74.2% | 14 | 31 | 56.7% | C | |
| SWE-Bench Pro | 59.0% | 14 | 45 | 70.5% | C | |
| BrowseComp | 83.5% | 16 | 58 | 73.7% | C | |
| OSWorld-Verified | 70.1% | 16 | 23 | 31.8% | C | |
| Terminal-Bench 2.1 | 66.0% | 16 | 19 | 16.7% | C | |
| MMMU-Pro | 78.1% | 21 | 66 | 69.2% | C | |
| LiveBench | 70.0% | 31 | 38 | 18.9% | B |
MiniMax M3 Arena Results
Preference and agent-evaluation results for the default version.
| agent bash recovery steps | overall | 22 | 0.1 | N/A | 50.4K | |
| document | overall | 26 | 1434.8 | 6,263 | N/A | |
| agent tool hallucination | overall | 29 | 0.0 | N/A | 1.6M | |
| webdev | overall | 31 | 1491.3 | 9,669 | N/A | |
| document style control | overall | 32 | 1433.5 | 6,263 | N/A | |
| agent | overall | 33 | -0.0 | N/A | 1.7M | |
| agent steerability | overall | 33 | -0.0 | N/A | 27.4K | |
| agent praise complaint | overall | 36 | -0.1 | N/A | 6.8K | |
| agent task outcome explicit | overall | 36 | -0.1 | N/A | 17.6K | |
| vision | overall | 42 | 1257.7 | 10,984 | N/A | |
| vision style control | overall | 47 | 1239.7 | 10,984 | N/A | |
| text | overall | 71 | 1435.2 | 36,687 | N/A | |
| text style control | overall | 71 | 1443.2 | 36,687 | N/A | |
| text factuality | overall | 83 | 1435.4 | 36,669 | N/A |
MiniMax M3 Runtime Performance
Provider-specific output speed and catalog latency for MiniMax M3. Runtime does not affect the capability score.
| Fireworks | 452.211 tok/s | 0.534 s | 512K | 131.1K | |
| Together | 192.197 tok/s | 0.482 s | 1M | 131.1K | |
| MiniMax | 78.826 tok/s | 1.091 s | 1M | 1M | |
| Novita | 44.113 tok/s | 3.205 s | 1M | 131.1K |
Output Speed is generated output tokens received per second. Catalog latency is reported separately from observed provider TTFT.
MiniMax M3 Pricing
Official vendor API pricing appears first, followed by individual provider offers.
- Official API
- $0.30 input, $1.2 output per 1M
- Official provider
- MiniMax (minimax.io)
- Lowest third-party
- From $0.225 input, $0.90 output per 1M via EmpirioLabs AI
- Tracked offerings
- 33
| Nvidia | minimaxai/minimax-m3 | global | N/A | N/A | 1M | |
| Kenari | minimax-m3 | global | N/A | N/A | 512K | |
| MiniMax Token Plan (minimax.io) | MiniMax-M3 | global | N/A | N/A | 1M | |
| MiniMax Token Plan (minimaxi.com) | MiniMax-M3 | global | N/A | N/A | 1M | |
| EmpirioLabs AI | minimax-m3 | global | $0.225 | $0.90 | 1M | |
| Weights & Biases | MiniMaxAI/MiniMax-M3 | global | $0.23 | $0.96 | 262.1K | |
| Lilac | minimaxai/minimax-m3 | global | $0.28 | $1.1 | 1M | |
| Deep Infra | MiniMaxAI/MiniMax-M3 | global | $0.28 | $1.1 | 524.3K | |
| Fireworks AI | accounts/fireworks/models/minimax-m3 | global | $0.30 | $1.2 | 512K | |
| OpenCode Go | minimax-m3 | global | $0.30 | $1.2 | 1M | |
| Nebius Token Factory | MiniMaxAI/MiniMax-M3 | global | $0.30 | $1.2 | 1M | |
| NanoGPT | minimax/minimax-m3 | global | $0.30 | $1.2 | 512K | |
| LLM Gateway | minimax-m3 | global | $0.30 | $1.2 | 1M | |
| OpenCode Zen | minimax-m3 | global | $0.30 | $1.2 | 512K | |
| ClinePass | cline-pass/minimax-m3 | global | $0.30 | $1.2 | 512K | |
| Pioneer | MiniMaxAI/MiniMax-M3 | global | $0.30 | $1.2 | 1M | |
| Abacus | MiniMaxAI/MiniMax-M3 | global | $0.30 | $1.2 | 1M | |
| Together AI | MiniMaxAI/MiniMax-M3 | global | $0.30 | $1.2 | 524.3K | |
| MiniMax (minimax.io) | MiniMax-M3 | global | $0.30 | $1.2 | 1M | |
| Kilo Gateway | minimax/minimax-m3 | global | $0.30 | $1.2 | 524.3K | |
| Merge Gateway | minimax/minimax-m3 | global | $0.30 | $1.2 | 1M | |
| Vercel AI Gateway | minimax/minimax-m3 | global | $0.30 | $1.2 | 1M | |
| OpenRouter | minimax/minimax-m3 | global | $0.30 | $1.2 | 1M | |
| Hugging Face | MiniMaxAI/MiniMax-M3 | global | $0.30 | $1.2 | 524.3K | |
| MiniMax (minimaxi.com) | MiniMax-M3 | global | $0.30 | $1.2 | 1M | |
| Charm Hyper | minimax-m3 | global | $0.3266 | $1.31 | 512K | |
| CrossModel | minimax/minimax-m3 | global | $0.33 | $1.32 | 1M | |
| Wafer | MiniMax-M3 | global | $0.33 | $1.32 | 1M | |
| Cortecs | minimax-m3 | global | $0.395 | $1.98 | 1M | |
| TensorX | minimax/minimax-m3 | global | $0.40 | $2 | 1M | |
| ZenMux | minimax/minimax-m3 | global | $0.60 | $2.4 | 512K | |
| Synthetic | hf:MiniMaxAI/MiniMax-M3 | global | $0.60 | $1.2 | 524.3K |
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
MiniMax M3 Versions
Available versions of this model. The score column identifies the version used in the overall ranking.
| MiniMax M3 | 71.3 | 428B | 512K | 128K | Yes | MIT |
What is MiniMax M3?
Key information about MiniMax M3 and its available data.
MiniMax M3 is the first open-weight model to combine three frontier capabilities: top-tier coding and agentic performance, a 1M-token context window, and native multimodality.
It is powered by MiniMax Sparse Attention (MSA), a new sparse attention architecture that partitions the KV cache into blocks to cut per-token compute at long context — roughly 1/20 the cost of the previous generation at 1M tokens, with more than 9x faster prefill and more than 15x faster decode while matching full attention on most capabilities.
Trained with mixed-modality data from step zero across 100T+ tokens, M3 natively supports image and video input and can operate a desktop computer. 7. M3 supports toggling thinking on or off at request time.
Data as of 2026-08-11.
MiniMax M3 vs nearby models
Open a comparison with the three ranked models immediately above and below this model.
Models similar to MiniMax M3
Recommendations prioritize the same model type and family, then the closest LLMBoard score.
MI
MiniMax M2.7
MiniMax
MI
MiniMax M2.5
MiniMax
MI
MiniMax M2.1
MiniMax
MI
MiniMax M2
MiniMax
MI
MiniMax M1 80K
MiniMax
MI
MiniMax M1 40K
MiniMax
FAQ
Common questions about MiniMax M3.
When was MiniMax M3 released?
MiniMax M3's default version was released on Jun 1, 2026.
How much does MiniMax M3 cost?
MiniMax M3's official API price is $0.30 per million input tokens and $1.2 per million output tokens via MiniMax (minimax.io). The lowest tracked third-party offer starts at $0.225 input and $0.90 output via EmpirioLabs AI.
Who created MiniMax M3?
MiniMax M3 was created by MiniMax.
What is the context window for MiniMax M3?
The default version has a 512K token context window.
Is MiniMax M3 open weight?
Yes. The default version is marked as open weight under MIT.
How many API providers offer MiniMax M3?
33 provider offerings are linked to the default version.
What models should I compare MiniMax M3 with?
Nearby ranked alternatives include Claude Sonnet 4.6, GPT-5.2-Pro, GPT-5.2.