MODEL · MINIMAX · ~229B TOTAL / ~10B ACTIVE (MOE, INTERLEAVED THINKING)
MiniMax M2.5 / M2.7
MiniMax's open-weights agentic-workflow family — strong on coding + tool-use Arena. M2.7 is the first model that "participates in its own evolution" via self-iterated RL. Frontier-class, not a local workhorse.
License: M2.5: MIT · M2.7: non-commercial with mandatory attribution · Context: 200K · Released: M2.5: February 12, 2026 · M2.7: March 18, 2026
The decision in five lines
- The call
- Hosted only
- Best for
- Hosted reference and benchmarks
- Runs on
- Hosted or workstation-class only · ~120 GB+ (not consumer-local)
- Watch out
- Anything with less than a multi-GPU or dual-3090 rig — these are cloud-or-serious-local only.
- Evidence
- Estimated
- ~229B total
- PARAMETERS
- MOE
- TYPE
- 200K
- CONTEXT
- ~120 GB+ (not consumer-local)
- VRAM AT Q4
Where we recommend this
This model isn’t currently in an active planner slot. See the runner notes below if you’re running it anyway.
The call
MiniMax's open-weights agentic-workflow family — strong on coding + tool-use Arena. M2.7 is the first model that "participates in its own evolution" via self-iterated RL. Frontier-class, not a local workhorse.
When not to use: Anything with less than a multi-GPU or dual-3090 rig — these are cloud-or-serious-local only. Also skip M2.7 in anything commercial without signing MiniMax's commercial license (the "Built with MiniMax M2.7" attribution is mandatory).
Runner notes
Realistically hosted via MiniMax API, Together, or OpenRouter. Prefer M2.5 for commercial work — MIT license, no attribution strings attached.
Next step
Find-by-model — see what hardware runs this→