MiniMax M2.7
MiniMax · released Apr 9, 2026 · MiniMaxAI/MiniMax-M2.7
- Type
- Open weightsCustom licence
- Params
- 229B
- Context
- 205K
about 154K words of context · download allowed, licence restricts use
Our take
Written Aug 3, 2026MiniMax M2.7 is a dense text model with 228.7 billion parameters and a 204,800-token request limit, released in April 2026 under a custom restricted licence. Its measured coding score is well above its general chat and creative writing scores, and it is available from eight different providers.
Choose this for coding tasks where the Arena Coding Elo score is the relevant signal, or for long-context text work up to 204,800 tokens. It suits budget-conscious inference through the cheapest host. Skip it if you need permissive licensing, consistent throughput across providers, or strong creative writing output.
The case for it
- Strongest measured skill is coding, with an Arena Coding Elo 62.6 points above its own overall text score.
- Ten offers from eight providers, with the cheapest rate 20% below the most common price point.
- Substantial request limit for a dense model of this scale, at 204,800 tokens.
The case against it
- Creative writing is a clear weak point, with an Elo 113.9 points below its coding score and 51.4 points below its overall text score.
- Throughput varies more than tenfold across providers for the same model, from 42 to 4 tokens per second.
- Custom restricted licence, not Apache or MIT; full 228.7 billion parameters are active per token with no disclosed efficiency architecture.
How good is it?
IntelligencePuzzles, maths, exam questions
Arena Text (overall)61st of 143 · 1416.9
CodingWriting and fixing code on its own
Arena Coding51st of 143 · 1479.4
AgenticPlanning, calling tools, staying on task
Arena Agent (IPS)33rd of 36 · −0.116
Arena Agent (IPS) is the only board that has scored it for this.
WritingWe do not rate this
Two boards come close and neither tests writing: Arena Creative Writing asks people which of two replies they prefer, and LiveBench Language tests whether a model understood a passage. So we show where MiniMax M2.7 placed and give it no mark out of five.
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, which is why they get no rating.
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?
- Fits in memory
- weights load entirely on the card
- Spills to system RAM
- some weights offload; much slower
- Too large
- will not load even with offload
- est
- size is calculated; the verdict could change by 10%
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.
All 3 sizes here are calculated, not measured. We hold no measured file for this model, so each size comes from the parameter count and every verdict above inherits that uncertainty.
Check against your own machine → · All 71 devices, with every size →
Or rent it from someone else
Cheapest of 18 live listings. Picked at the widest standard context we hold, within one quantisation slice, so the numbers beside it are a price one host actually charges.
- per 1M tokens
- $0.25 in / $1.00 out
- Context served
- 205K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| Mara | $0.24 / $0.96 | 197K | 39 tok/s | No | No | Confirmed |
| DeepInfrafp8 | $0.25 / $1.00 | 197K | 9 tok/s | No | No | Confirmed |
| DeepInfrafp8 | $0.25 / $1.00 | 197K | not measured | Unknown | Unknown | Unknown |
| OpenRouter | $0.25 / $1.00 | 205K | not measured | Unknown | Unknown | Unknown |
| Novita AIfp8 | $0.27 / $1.08 | 205K | 31 tok/s | No | No | Confirmed |
| Novita AI | $0.30 / $1.20 | 205K | not measured | Unknown | Unknown | Unknown |
| GMICloudfp8 | $0.30 / $1.20 | 197K | 37 tok/s | No | Yesunknown period | Unknown |
| Together AIfp4 | $0.30 / $1.20 | 197K | not measured | No | No | Unknown |
| AtlasCloudfp8 | $0.30 / $1.20 | 197K | 27 tok/s | No | Yesunknown period | Unknown |
| Minimaxfp8 | $0.30 / $1.20 | 205K | 44 tok/s | No | Yesunknown period | Confirmed |
| Fireworks AI | $0.30 / $1.20 | 197K | 174 tok/s | No | No | Confirmed |
| SambaNovaminimax-m2.7-dedicated | $0.30 / $1.50 | 197K | 165 tok/s | No | No | Unknown |
| SambaNova | $0.30 / $1.50 | 197K | 196 tok/s | No | No | Confirmed |
| NextBitfp8 | $0.40 / $1.60 | 197K | 76 tok/s | No | No | Unknown |
| DeepInfraturbo tierfp8 | $0.38 / $1.70 | 197K | 56 tok/s | No | No | Unknown |
| Groq | $0.60 / $1.80 | 197K | 386 tok/s | No | No | Confirmed |
| SambaNova | $0.60 / $2.40 | 197K | not measured | Unknown | Unknown | Unknown |
| Minimaxhighspeed tierfp8 | $0.60 / $2.40 | 205K | 59 tok/s | No | Yesunknown period | Unknown |
DeepInfra
Novita AI
Together AI
Fireworks AI
SambaNova
GroqAcross the 18 listings we hold: 14 say they do not train on prompts, 0 say they do and 4 do not say. 7 appear in the zero-retention registry we check; the rest are unknown to us rather than confirmed either way.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
- ✓
- Supported
- ✗
- Not supported
- Not published
- host gave no parameter list
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| Mara | ✓ | ✓ | ✓ |
| DeepInfrafp8 | ✓ | ✓ | ✗ |
| DeepInfrafp8 | |||
| OpenRouter | ✓ | ✓ | ✓ |
| Novita AIfp8 | ✓ | ✓ | ✗ |
| Novita AI | |||
| GMICloudfp8 | ✓ | ✓ | ✗ |
| Together AIfp4 | |||
| AtlasCloudfp8 | ✓ | ✓ | ✗ |
| Minimaxfp8 | ✓ | ✓ | ✗ |
| Fireworks AI | ✓ | ✓ | ✓ |
| SambaNovaminimax-m2.7-dedicated | ✓ | ✗ | ✓ |
| SambaNova | ✓ | ✗ | ✓ |
| NextBitfp8 | ✓ | ✓ | ✓ |
| DeepInfraturbo · fp8 | ✓ | ✓ | ✗ |
| Groq | ✓ | ✗ | ✗ |
| SambaNova | |||
| Minimaxhighspeed · fp8 | ✓ | ✓ | ✗ |
Tool calling: 14 of 18 listings say yes, 4 publish no parameter list. JSON output: 11 of 18 listings say yes, 3 say no, 4 publish no parameter list. Strict schema: 6 of 18 listings say yes, 8 say no, 4 publish no parameter list.
Models people weigh against MiniMax M2.7
When we formed this view
Dates behind this page
Prices last checked 3d ago
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.
- 4 of 18 listings publish no parameter list, so what their API accepts is unknown to us.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 4 of 18 listings do not say whether they train on prompts.
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-M2.7
- Architecture
- Mixture of experts
- Modality record
- text->text
- Catalogue slug
- minimax-minimax-m2-7