MiniMax-M3
Canonical page on the main site: sinoaihub.com/models/minimax-m3
Type
model
Definition
MiniMax-M3 is a MiniMax model in the MiniMax M-series family: Mixture-of-Experts: ~428B total / ~23B activated; MiniMax Sparse Attention (MSA) with claimed 9x prefill and 15x decode speedup vs M2 at 1M context; 1,048,576-token context window; 131,072 max output; open-weight; MiniMax Community License (custom): free for non-commercial use; commercial use requires prominent attribution 'Built with MiniMax M3' plus written authorization from MiniMax if yearly revenue exceeds US$20M (otherwise a one-time notice to [email protected]) license; released 2026-06-01.
Key facts
- Architecture: Mixture-of-Experts: ~428B total / ~23B activated; MiniMax Sparse Attention (MSA) with claimed 9x prefill and 15x decode speedup vs M2 at 1M context
- Context window: 1,048,576 tokens (max output 131,072)
- Weights: open; MiniMax Community License (custom): free for non-commercial use; commercial use requires prominent attribution 'Built with MiniMax M3' plus written authorization from MiniMax if yearly revenue exceeds US$20M (otherwise a one-time notice to [email protected]) license
- API pricing: $0.3 input / $1.2 output per 1M tokens (USD)
- Released: 2026-06-01
- Capabilities: reasoning, coding, vision, tool calling
Provider
Model Family
MiniMax M-series
Release Date
2026-06-01
Status
active
Architecture
Mixture-of-Experts: ~428B total / ~23B activated; MiniMax Sparse Attention (MSA) with claimed 9x prefill and 15x decode speedup vs M2 at 1M context
Parameter Information
- total_parameters: ~428B
- active_parameters: ~23B
Context Window
1048576
Capabilities
- reasoning: Yes
- coding: Yes
- vision: Yes
- video: Yes
- tool_calling: Yes
- agent_capability: Yes
Open Weight
Yes
License
MiniMax Community License (custom): free for non-commercial use; commercial use requires prominent attribution 'Built with MiniMax M3' plus written authorization from MiniMax if yearly revenue exceeds US$20M (otherwise a one-time notice to [email protected])
Self Hosting
Yes
Api Available
Yes
Pricing
- input_price_per_1m: 0.3
- output_price_per_1m: 1.2
- currency: USD
- pricing_ref: minimax
Official Api
Yes
Cloud Providers
- MiniMax Platform
Regions
- china
- international
Benchmark Results
- BrowseComp: 83.5 (vendor_reported)
- PostTrainBench: 37.1 (vendor_reported)
- SWE-bench Pro: 59.0 (vendor_reported)
- Terminal-Bench 2.1: 66.0 (vendor_reported)
- MCP Atlas: 74.2 (vendor_reported)
Related Agents
Related Technologies
Maximum Output
131072
Known Limitations
- Maximum output of 131,072 tokens derived from official card evaluation config (128K max output tokens); the standalone API max-output ceiling is not separately published
- Max output tokens not publicly disclosed in official docs
- 1M context with at least 512K guaranteed usable; pricing splits at the 512K input boundary
- Thinking is disabled by default (thinking=adaptive enables it)
- Benchmarks vendor-reported; not independently verified
Verification Status
partially_verified
License Changes
No documented license changes on record as of 2026-09-29.
Last Verified
2026-09-20
Source history
No documented source-change events located as of 2026-09-20.
Sources
| evidence_id | source_name | source_url | source_type | published | verified | confidence | conflict |
|---|---|---|---|---|---|---|---|
| src-models-minimax-m3-1 | MiniMax API platform — model overview (CN) | https://platform.minimaxi.com/docs/guides/models-intro | Official documentation | — | 2026-09-20 | high | — |
| src-models-minimax-m3-2 | MiniMax official M3 model page | https://www.minimax.cn/models/text/m3 | Official documentation | — | 2026-09-20 | high | — |
| src-models-minimax-m3-3 | MiniMax M3 official blog post | https://www.minimax.cn/blog/minimax-m3 | Official | 2026-06-01 | 2026-09-20 | high | — |
| src-models-minimax-m3-4 | Hugging Face model card — MiniMax-M3 | https://huggingface.co/MiniMaxAI/MiniMax-M3 | Model card | — | 2026-09-20 | high | — |