LLMBenchmark.ai

Enter opens the highlighted result. Esc closes.

MiniMax M2.5 vs MiniMax M2.7

Updated October 21, 2026. 59 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.

Replace MiniMax M2.5 Replace MiniMax M2.7
MiniMax M2.5MiniMax
Coding
59.4
Overall
22.8
Output $/M
$0.95
9
MiniMax M2.7MiniMax
Coding
28.4
Overall
22.8
Output $/M
$1.20
49
vs boards won of 59 shared, 1 tied

Decision reading

Pick MiniMax M2.5 for coding: it scores 59.4 on the Vibe Coding Index against 28.4 for MiniMax M2.7.

Pick MiniMax M2.7 when what matters is LMArena WebDev or LiveCodeBench (Vals AI), where it leads.

Model AMiniMax M2.5MiniMax59.4/1002 independent benchmarks$0.62 per solved taskModel BMiniMax M2.7MiniMax28.4/1001 independent benchmark$0.64 per solved task
Shared results
59
MiniMax M2.5 only
8
MiniMax M2.7 only
15
Like-for-like categories
8 of 9

Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.

Which one for your work

Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.

At a glance

Coding index
MiniMax M2.5 59.4 from 2 benchmarks, MiniMax M2.7 28.4 from 1 benchmark.
Widest category
Agentic work, 10.5 points apart on average over 1 shared benchmark.
Widest single gap
LMArena Text: Japanese: MiniMax M2.5 1171 against MiniMax M2.7 1260, 14.8 points apart on a 0 to 100 scale.
Output price
MiniMax M2.5 $0.95 against MiniMax M2.7 $1.20 per million output tokens, 1.3x apart.
Context window
Both take 205K tokens.

Benchmark by benchmark

First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.

Boards won, where both have a result
CategoryMiniMax M2.5BarsMiniMax M2.7Ties
Coding47
Intelligence101 tie
Agentic work10
Reasoning06
Math13
Knowledge18
Web design16
General019
All shared boards9491 tie
CodingLike-for-likeFour of these boards feed the Vibe Coding Index47
Coding benchmarks for MiniMax M2.5 and MiniMax M2.7
BenchmarkMiniMax M2.5MiniMax M2.7Leader
SWE-bench Verified75.8%Not measuredOnly MiniMax M2.5 measured
SWE-bench Multilingual68.3%Not measuredOnly MiniMax M2.5 measured
LMArena WebDev13861399MiniMax M2.7
AA Coding IndexNot measured52.6Only MiniMax M2.7 measured
LiveCodeBench (Vals AI)79.2%79.9%MiniMax M2.7
SWE-bench Verified (Vals AI)74.2%73.8%MiniMax M2.5
Vibe Code Bench (Vals AI)14.8%11.9%MiniMax M2.5
Code migration (Vals AI)Not measured8.7%Only MiniMax M2.7 measured
ProgramBench (Vals AI)Not measured0.0%Only MiniMax M2.7 measured
Terminal-Bench 2.1 (Vals AI)Not measured48.7%Only MiniMax M2.7 measured
SciCodeNot measured50.1%Only MiniMax M2.7 measured
Terminal-Bench Hard (Artificial Analysis run)34.8%39.4%MiniMax M2.7
Terminal-Bench42.7%45.1%MiniMax M2.7
WeirdMLNot measured37.0%Only MiniMax M2.7 measured
ALE-Bench618.2599.3MiniMax M2.5
LMArena Text: coding13801454MiniMax M2.7
LMArena Text: software and IT13831446MiniMax M2.7
LMArena WebDev: React13741388MiniMax M2.7
LMArena WebDev: HTML14141401MiniMax M2.5
Neither model has a result on the other 19 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index10
Intelligence benchmarks for MiniMax M2.5 and MiniMax M2.7
BenchmarkMiniMax M2.5MiniMax M2.7Leader
AA Intelligence Index22.822.8Tie
Epoch Capabilities Index146.6145.8MiniMax M2.5
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
Agentic work benchmarks for MiniMax M2.5 and MiniMax M2.7
BenchmarkMiniMax M2.5MiniMax M2.7Leader
AA Agentic IndexNot measured15.3Only MiniMax M2.7 measured
Finance Agent (Vals AI)Not measured27.9%Only MiniMax M2.7 measured
Tax Agent Bench (Vals AI)Not measured6.9%Only MiniMax M2.7 measured
Legal Research Bench (Vals AI)Not measured10.6%Only MiniMax M2.7 measured
Excel Modeling Benchmark (Vals AI)Not measured28.4%Only MiniMax M2.7 measured
τ²-Bench Telecom95.3%84.8%MiniMax M2.5
Vending-Bench 2-$23Not measuredOnly MiniMax M2.5 measured
Neither model has a result on the other 32 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index06
Reasoning benchmarks for MiniMax M2.5 and MiniMax M2.7
BenchmarkMiniMax M2.5MiniMax M2.7Leader
GPQA Diamond82.1%86.6%MiniMax M2.7
GPQA Diamond (Artificial Analysis run)84.8%87.4%MiniMax M2.7
Humanity's Last Exam (Artificial Analysis run)20.5%29.6%MiniMax M2.7
ARC-AGI63.7%Not measuredOnly MiniMax M2.5 measured
ARC-AGI-24.9%Not measuredOnly MiniMax M2.5 measured
LMArena Text: hard prompts13721422MiniMax M2.7
LMArena Text: expert13781443MiniMax M2.7
LMArena Text: hard prompts in English13841441MiniMax M2.7
Neither model has a result on the other 13 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index13
Math benchmarks for MiniMax M2.5 and MiniMax M2.7
BenchmarkMiniMax M2.5MiniMax M2.7Leader
AIME88.8%91.0%MiniMax M2.7
ProofBench4.0%3.0%MiniMax M2.5
LMArena Text: math13781421MiniMax M2.7
LMArena Text: mathematical work13781437MiniMax M2.7
Neither model has a result on the other 9 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index18
Knowledge benchmarks for MiniMax M2.5 and MiniMax M2.7
BenchmarkMiniMax M2.5MiniMax M2.7Leader
MMLU Pro80.1%80.4%MiniMax M2.7
LegalBench (Vals AI)80.0%84.0%MiniMax M2.7
CaseLaw (Vals AI)53.5%60.9%MiniMax M2.7
CorpFin (Vals AI)59.6%61.2%MiniMax M2.7
TaxEval (Vals AI)68.2%66.6%MiniMax M2.5
MedQA (Vals AI)92.5%Not measuredOnly MiniMax M2.5 measured
MedCode (Vals AI)Not measured34.4%Only MiniMax M2.7 measured
LMArena Text: science13691416MiniMax M2.7
LMArena Text: medicine and healthcare13571409MiniMax M2.7
LMArena Text: legal and government13661403MiniMax M2.7
LMArena Text: business and finance13561407MiniMax M2.7
Neither model has a result on the other 6 boards here
Web designLike-for-likePublic boards, shown as published, kept out of the index16
Web design benchmarks for MiniMax M2.5 and MiniMax M2.7
BenchmarkMiniMax M2.5MiniMax M2.7Leader
Design Arena: websites12221243MiniMax M2.7
Design Arena: UI components11721210MiniMax M2.7
Design Arena: game dev11761210MiniMax M2.7
Design Arena: data viz11711233MiniMax M2.7
Design Arena: 3D11701195MiniMax M2.7
Design Arena: all code categories12071233MiniMax M2.7
Design Arena: SVG11471136MiniMax M2.5
Design Arena: ASCII artNot measured1136Only MiniMax M2.7 measured
Neither model has a result on the other 1 board here
GeneralLike-for-likePublic boards, shown as published, kept out of the index019
General benchmarks for MiniMax M2.5 and MiniMax M2.7
BenchmarkMiniMax M2.5MiniMax M2.7Leader
MedScribe (Vals AI)Not measured79.9%Only MiniMax M2.7 measured
IFBench71.6%75.7%MiniMax M2.7
Long context reasoning (AA-LCR)73.3%78.3%MiniMax M2.7
CL-bench11.4%Not measuredOnly MiniMax M2.5 measured
ExploitBenchNot measured13.3%Only MiniMax M2.7 measured
CL-bench Life6.3%Not measuredOnly MiniMax M2.5 measured
LMArena Text13591405MiniMax M2.7
LMArena Text: instruction following13531405MiniMax M2.7
LMArena Text: creative writing13311354MiniMax M2.7
LMArena Text: English13801426MiniMax M2.7
LMArena Text: non-English13381382MiniMax M2.7
LMArena Text: Chinese13931442MiniMax M2.7
LMArena Text: German13611398MiniMax M2.7
LMArena Text: Japanese11711260MiniMax M2.7
LMArena Text: Korean12331314MiniMax M2.7
LMArena Text: French13631424MiniMax M2.7
LMArena Text: Spanish13531402MiniMax M2.7
LMArena Text: Russian13591383MiniMax M2.7
LMArena Text: Polish13341387MiniMax M2.7
LMArena Text: multi-turn13641412MiniMax M2.7
LMArena Text: longer queries13661418MiniMax M2.7
LMArena Text: writing and language13451376MiniMax M2.7
LMArena Text: entertainment and media13291360MiniMax M2.7
Neither model has a result on the other 23 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for MiniMax M2.5 and MiniMax M2.7
BenchmarkMiniMax M2.5MiniMax M2.7Leader
Neither model has a result on 5 boards here

What each workload costs

Three fixed token mixes at each model's listed price, so per-token rates become a bill.

Chat turn

1K input + 500 output tokens

MiniMax M2.5$0.0007Fits in one request
MiniMax M2.7$0.0009Fits in one request

MiniMax M2.5 costs less

Repository review

50K input + 3K output tokens

MiniMax M2.5$0.016Fits in one request
MiniMax M2.7$0.019Fits in one request

MiniMax M2.5 costs less

Agent loop

200K input + 10K output tokens

MiniMax M2.5$0.064Exceeds the context window
MiniMax M2.7$0.072Exceeds the context window

MiniMax M2.5 costs less

Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.

Specification differences

The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.

Specifications of MiniMax M2.5 and MiniMax M2.7
SpecificationMiniMax M2.5MiniMax M2.7
MakerMiniMaxMiniMax
WeightsOpen weightsOpen weights
ReleasedFebruary 12, 2026March 18, 2026
Context window205K205K
Input $/Mtok$0.27$0.30
Output $/Mtok$0.95$1.20
Coding index59.428.4
Independent measurements behind it21
Cost per solved task$0.62$0.64
SourceOpenRouter listingOpenRouter listing

Questions

Is MiniMax M2.5 better than MiniMax M2.7 for coding?

Pick MiniMax M2.5 for coding: it scores 59.4 on the Vibe Coding Index against 28.4 for MiniMax M2.7. Pick MiniMax M2.7 when what matters is LMArena WebDev or LiveCodeBench (Vals AI), where it leads.

Which is cheaper, MiniMax M2.5 or MiniMax M2.7?

MiniMax M2.5 costs $0.27 per million input tokens and $0.95 per million output tokens; MiniMax M2.7 costs $0.30 and $1.20. Per solved SWE-bench Verified task, MiniMax M2.5 costs $0.62 and MiniMax M2.7 $0.64.

Which has the bigger context window, MiniMax M2.5 or MiniMax M2.7?

Neither: both take 205K tokens.

How many benchmarks is the MiniMax M2.5 vs MiniMax M2.7 comparison based on?

59 benchmarks have a result for both models as of October 21, 2026. MiniMax M2.5's coding index rests on 2 independent benchmarks and MiniMax M2.7's on 1. Where only one model has a result, the page says so instead of naming a winner.

Compare next