Developers wanting million-token context for coding agents at low prices should look, though the price relies on a 50% discount.
Kenning's take
What it does
MiniMax M3 is an open-weight large language model built for coding and long agent tasks. A new sparse attention design lets the API take up to one million tokens of context, and it was trained on text and images from the start so it can read charts, screenshots and video. It is available through the MiniMax API and M Plan subscription, in MiniMax Code, and as weights on Hugging Face.
- Up to 1M token context (512K guaranteed)
- MiniMax Sparse Attention architecture
- Native image and video understanding
- 83.5 on BrowseComp (MiniMax figure)
- Automatic prompt caching in the API
Who it suits. Developers running long coding and research agents who want an open model with a very long context.
Key facts
- Category
- AI models
- Made by
- MiniMax
- Based in
- China
- Released
- 1 June 2026
- How to get it
- through the MiniMax API (pay as you go or M Plan), in MiniMax Code, and as open weights on Hugging Face under the MiniMax community licence
- Works on
- API, self-hosted
- Free plan
- Open weights are a free download
- API price
- Up to 512K input: $0.30 per million input tokens, $1.20 per million output (list $0.60 and $2.40, 'permanent 50% off'); above 512K: $0.60 input, $2.40 output
Reviews and write-ups
- The Decoder 2026-06-01
FAQ
When was MiniMax M3 released?
MiniMax released M3 on 1 June 2026, according to The Decoder.
How long is MiniMax M3's context window?
Up to one million tokens in the API, with at least 512K guaranteed.
Is MiniMax M3 open source?
Its weights are on Hugging Face under the MiniMax community licence.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.