MiniMax M2.5
MiniMax · released Feb 12, 2026 · MiniMaxAI/MiniMax-M2.5
- Type
- Open weightsCustom licence
- Params
- 229B
- Context
- 205K
10B active per word · about 154K words of context · download allowed, licence restricts use
Our take
Written Sep 17, 2026MiniMax M2.5 is a downloadable text model with a custom licence that puts conditions on commercial use and redistribution, so it needs reading before you build on it. Its strongest measured result is on real GitHub issue fixing, and many hosts serve it.
Use it for fixing real GitHub issues where you can run the mini-SWE-agent harness, since that is the only end-to-end coding measurement we hold, or for everyday chat, writing and instruction-following work where human preference is a reasonable proxy for quality. Skip it if you need a licence that allows commercial use without conditions, or if you need measured speed on any host.
The case for it
- 75.8% of real GitHub issues resolved end-to-end on SWE-bench Verified, measured inside the mini-SWE-agent harness, so the figure describes the model in that harness rather than on its own.
- Arena scores across coding, creative writing, hard prompts, instruction following, maths, overall text and web-app building, all from human pairwise votes, which record which answer people preferred rather than whether it was correct.
- The request capacity is large enough for long documents to be handled in one go, though reliable recall across all of it is unverified in our data.
The case against it
- The licence puts conditions on commercial use and redistribution, so it needs reading before you build on it (Custom licence).
- No tokens-per-second figure is supplied for any of the 18 hosted offers, so speed cannot be compared between them.
- The only non-Arena score is 75.8% on SWE-bench Verified inside the mini-SWE-agent harness; nothing supplied measures standalone coding, reasoning or maths as a percentage.
How good is it?
EverydayGeneral questions and everyday reasoning
Arena Text (overall)104th of 168 · 1391
CodingWriting and fixing code on its own
Arena Coding97th of 168 · 1446
AgenticPlanning, calling tools, staying on task
Not yet scored on Arena Agent. It is on SWE-bench Verified, in 2nd of 42 with 75.8.
WritingDrafting and rewriting prose
Arena Creative Writing98th of 168 · 1357
Arena Creative Writing is the only board that has scored it for this.
These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done.
Every published score for this model8 scoresEvery figure we hold, from 8 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
GeForce RTX 4090 · 24 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Apple M1 Pro (16-core GPU) · 32 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Comfortable fit
Apple M3 Ultra (80-core GPU) · 512 GB
Room to spare. 233.8 GB spare means a 10% error in the size would not change the answer.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first.
Check against your own machine → · Where to rent it hosted →
Or rent it from someone else
Prices checked 1 hour ago — each listing carries its own date.
The only listing at 205K of context — the other 8 in the table below are not like-for-like. One cheaper row there is outside that comparison: a different context length.
- per 1M tokens
- $0.27 in / $1.08 out
- Context served
- 205K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| Venice AIThrough OpenRouter | $0.27 / $0.95checked 1 hour ago | 198K33K max reply | 19 tok/s | No | No | Confirmed |
| StreamLakeThrough OpenRouter | $0.27 / $1.08checked 1 hour ago | 200K128K max reply | 53 tok/s | No | Yesunknown period | Unknown |
| OpenRouterOpenRouter's own listing | $0.27 / $1.08checked 1 hour ago | 205K | not measured | Unknown | Unknown | Unknown |
| AtlasCloudfp8Through OpenRouter | $0.29 / $1.20checked 1 hour ago | 197K177K max reply | 61 tok/s | No | Yesunknown period | Unknown |
| DigitalOcean GradientThrough OpenRouter | $0.30 / $1.20checked 1 hour ago | 66K59K max reply | 57 tok/s | No | No | Confirmed |
| Minimaxfp8Through OpenRouter | $0.30 / $1.20checked 1 hour ago | 205K131K max reply | 64 tok/s | No | Yesunknown period | Confirmed |
| FriendliThrough OpenRouter | $0.30 / $1.20checked 1 hour ago | 197K177K max reply | 110 tok/s | No | Yesunknown period | Unknown |
| Novita AIfp8Direct and through OpenRouter | $0.30 / $1.20checked 1 hour ago | 205K131K max reply through OpenRouter | 60 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| Minimaxhighspeed tierfp8Through OpenRouter | $0.60 / $2.40checked 1 hour ago | 205K131K max reply | 36 tok/s | No | Yesunknown period | Unknown |
DigitalOcean Gradient
Novita AIAcross the 9 listings we hold: 8 say they do not train on prompts (1 of them only through OpenRouter), 0 say they do and 1 does not say. 4 appear in the zero-retention registry we check (1 of them only through OpenRouter); the rest are unknown to us.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| Venice AIThrough OpenRouter | ✓ | ✗ | ✗ |
| StreamLakeThrough OpenRouter | ✓ | ✓ | ✓ |
| OpenRouterOpenRouter's own listing | ✓ | ✓ | ✓ |
| AtlasCloudfp8Through OpenRouter | ✓ | ✓ | ✓ |
| DigitalOcean GradientThrough OpenRouter | ✓ | ✓ | ✓ |
| Minimaxfp8Through OpenRouter | ✓ | ✓ | ✗ |
| FriendliThrough OpenRouter | ✓ | ✓ | ✓ |
| Novita AIfp8Direct and through OpenRouter | ✓ | ✓ | ✗ |
| Minimaxhighspeed · fp8Through OpenRouter | ✓ | ✓ | ✗ |
Tool calling: 9 of 9 listings say yes. JSON output: 8 of 9 listings say yes, 1 says no. Strict schema: 5 of 9 listings say yes, 4 say no.
Models people weigh against MiniMax M2.5
When we formed this view
Recent changes
What moved
input +33% ($0.225 → $0.300 per 1M tokens), output +33% ($0.90 → $1.20 per 1M tokens)What moved
input +47% ($0.15 → $0.22 per 1M tokens)What moved
input −50% ($0.60 → $0.30 per 1M tokens), output −50% ($2.40 → $1.20 per 1M tokens), cache read −50% ($0.060 → $0.030 per 1M tokens)Each date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 1 of 9 listings does not say whether it trains on prompts, and 1 answers only through OpenRouter, not for its own listing.
- We hold no batch or off-peak rate for any of its listings.
- We hold a decode speed for it, but no prompt-processing (prefill) figure, so how long the input side of a job takes is unknown to us.
Licence and identifiers
What the licence allowsCustom licence, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Custom licence
This model ships custom license terms that don't map to a known template. We haven't parsed them, so commercial use, redistribution and derivatives are unverified — review the original terms before shipping.
Identifiers
- Hugging Face
- MiniMaxAI/MiniMax-M2.5
- Architecture
- Mixture of experts
- Takes in, gives back
- Text in, text out
- Catalogue slug
- minimax-minimax-m2-5