Cheap per token and open to download, it appeals to developers building agents on a budget; UK availability is not stated.
Kenning's take
What it does
MiniMax M2.7 is a large language model from MiniMax aimed at software engineering, agent workflows and office documents. MiniMax says an internal version built agent tools and ran parts of its own reinforcement learning experiments, which it calls early self-evolution, and reports 56.22% on SWE-Pro. It is available through the MiniMax API and as downloadable weights on Hugging Face.
- 56.22% on SWE-Pro and 57.0% on Terminal Bench 2 (MiniMax figures)
- Agent Teams, complex skills and dynamic tool search
- Editing of Excel, PowerPoint and Word files
- Faster -highspeed API variant
- Open weights on Hugging Face
Who it suits. Developers building coding agents and multi-step automations on a low-cost model.
Key facts
- Category
- AI models
- Made by
- MiniMax
- Based in
- China
- Released
- 18 March 2026
- How to get it
- through the MiniMax API and coding plan, and as open weights on Hugging Face under MiniMax's own licence
- Works on
- API, self-hosted
- Free plan
- Open weights are a free download
- API price
- $0.30 per million input tokens, $1.20 per million output; highspeed variant $0.60 input, $2.40 output
FAQ
When was MiniMax M2.7 released?
MiniMax announced M2.7 on 18 March 2026.
What does self-evolving mean for M2.7?
MiniMax says an internal version of the model updated its own memory, built skills and helped run the reinforcement learning experiments used to improve it.
How much does the MiniMax M2.7 API cost?
$0.30 per million input tokens and $1.20 per million output tokens.
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.