LLMBenchmark.ai

Enter opens the highlighted result. Esc closes.

MiniMax M2.5 vs MiniMax M3

Updated October 21, 2026. 57 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 M3
MiniMax M2.5MiniMax
Coding
59.4
Overall
22.8
Output $/M
$0.95
1
MiniMax M3MiniMax
Coding
57.8
Overall
29.2
Output $/M
$1.20
56
vs boards won of 57 shared

Decision reading

MiniMax M2.5 and MiniMax M3 are level for coding, 59.4 against 57.8 on the Vibe Coding Index, so price and context decide.

Pick MiniMax M3 when a solved task costs $0.56 against $0.62, or when the context window is larger, 1.05M against 205K tokens. Pick MiniMax M2.5 when output tokens cost 1.3x less, or when the benchmark that matters is τ²-Bench Telecom, where it leads.

Model AMiniMax M2.5MiniMax59.4/1002 independent benchmarks$0.62 per solved taskModel BMiniMax M3MiniMax57.8/1002 independent benchmarks$0.56 per solved task
Shared results
57
MiniMax M2.5 only
10
MiniMax M3 only
59
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 M3 57.8 from 2 benchmarks.
Widest category
General, 14.1 points apart on average over 19 shared benchmarks.
Widest single gap
LMArena Text: Japanese: MiniMax M2.5 1171 against MiniMax M3 1380, 34.9 points apart on a 0 to 100 scale.
Output price
MiniMax M2.5 $0.95 against MiniMax M3 $1.20 per million output tokens, 1.3x apart.
Context window
MiniMax M2.5 takes 205K tokens, MiniMax M3 1.05M.

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 M3Ties
Coding010
Intelligence02
Agentic work11
Reasoning06
Math03
Knowledge08
Web design07
General019
All shared boards156
CodingLike-for-likeFour of these boards feed the Vibe Coding Index010
Coding benchmarks for MiniMax M2.5 and MiniMax M3
BenchmarkMiniMax M2.5MiniMax M3Leader
SWE-bench Verified75.8%Not measuredOnly MiniMax M2.5 measured
SWE-bench Multilingual68.3%Not measuredOnly MiniMax M2.5 measured
LiveBench codingNot measured70.9%Only MiniMax M3 measured
LMArena WebDev13861482MiniMax M3
AA Coding IndexNot measured58.6Only MiniMax M3 measured
LiveCodeBench (Vals AI)79.2%82.2%MiniMax M3
SWE-bench Verified (Vals AI)74.2%75.0%MiniMax M3
Terminal-Bench 4 (Vals AI)Not measured1.0%Only MiniMax M3 measured
Vibe Code Bench (Vals AI)14.8%47.6%MiniMax M3
Code migration (Vals AI)Not measured19.9%Only MiniMax M3 measured
Terminal-Bench 2.1 (Vals AI)Not measured53.6%Only MiniMax M3 measured
SWE-bench Multimodal (Vals AI)Not measured33.3%Only MiniMax M3 measured
Vibe Code Bench 1-100 (Vals AI)Not measured9.2%Only MiniMax M3 measured
SkillsBench (Vals AI)Not measured51.5%Only MiniMax M3 measured
CyberBench (Vals AI)Not measured65.0%Only MiniMax M3 measured
SciCodeNot measured47.1%Only MiniMax M3 measured
Terminal-Bench Hard (Artificial Analysis run)34.8%42.4%MiniMax M3
Terminal-Bench42.7%Not measuredOnly MiniMax M2.5 measured
FrontierCodeNot measured14.7%Only MiniMax M3 measured
ALE-Bench618.2640.0MiniMax M3
LMArena Text: coding13801470MiniMax M3
LMArena Text: software and IT13831466MiniMax M3
LMArena WebDev: React13741478MiniMax M3
LMArena WebDev: HTML14141455MiniMax M3
LMArena WebDev: from an imageNot measured1470Only MiniMax M3 measured
Neither model has a result on the other 13 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
Intelligence benchmarks for MiniMax M2.5 and MiniMax M3
BenchmarkMiniMax M2.5MiniMax M3Leader
AA Intelligence Index22.829.2MiniMax M3
Epoch Capabilities Index146.6146.9MiniMax M3
Agentic workLike-for-likePublic boards, shown as published, kept out of the index11
Agentic work benchmarks for MiniMax M2.5 and MiniMax M3
BenchmarkMiniMax M2.5MiniMax M3Leader
AA Agentic IndexNot measured29.5Only MiniMax M3 measured
Design Arena: agents building web appsNot measured1200Only MiniMax M3 measured
Design Arena: agents building full-stack appsNot measured1197Only MiniMax M3 measured
Design Arena: agents building mobile appsNot measured1169Only MiniMax M3 measured
Design Arena: agents building Android appsNot measured1164Only MiniMax M3 measured
Design Arena: agents building gamesNot measured1158Only MiniMax M3 measured
Design Arena: HTML slidesNot measured1180Only MiniMax M3 measured
Design Arena: PowerPoint slidesNot measured1205Only MiniMax M3 measured
Design Arena: spreadsheetsNot measured1021Only MiniMax M3 measured
LiveBench researchNot measured68.0%Only MiniMax M3 measured
LiveBench browserNot measured8.0%Only MiniMax M3 measured
LiveBench long agenticNot measured12.4%Only MiniMax M3 measured
Finance Agent (Vals AI)Not measured48.3%Only MiniMax M3 measured
Tax Agent Bench (Vals AI)Not measured20.6%Only MiniMax M3 measured
Legal Research Bench (Vals AI)Not measured29.8%Only MiniMax M3 measured
Excel Modeling Benchmark (Vals AI)Not measured47.8%Only MiniMax M3 measured
τ²-Bench Telecom95.3%88.9%MiniMax M2.5
APEX-AgentsNot measured37.7%Only MiniMax M3 measured
OSWorld 2.0Not measured4.6%Only MiniMax M3 measured
Vending-Bench 2-$23$2,158MiniMax M3
Neither model has a result on the other 19 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index06
Reasoning benchmarks for MiniMax M2.5 and MiniMax M3
BenchmarkMiniMax M2.5MiniMax M3Leader
GPQA Diamond82.1%92.7%MiniMax M3
LiveBench reasoningNot measured74.5%Only MiniMax M3 measured
MMMU Pro (Vals AI)Not measured81.2%Only MiniMax M3 measured
GPQA Diamond (Artificial Analysis run)84.8%92.9%MiniMax M3
Humanity's Last Exam (Artificial Analysis run)20.5%39.0%MiniMax M3
GPQA Diamond (Epoch AI run)Not measured90.9%Only MiniMax M3 measured
ARC-AGI63.7%Not measuredOnly MiniMax M2.5 measured
ARC-AGI-24.9%Not measuredOnly MiniMax M2.5 measured
SimpleBenchNot measured45.8%Only MiniMax M3 measured
Chess puzzlesNot measured13.0%Only MiniMax M3 measured
Mystery game puzzlesNot measured8.0%Only MiniMax M3 measured
LMArena Text: hard prompts13721447MiniMax M3
LMArena Text: expert13781461MiniMax M3
LMArena Text: hard prompts in English13841455MiniMax M3
LMArena Vision: diagramsNot measured1266Only MiniMax M3 measured
LMArena Vision: homeworkNot measured1268Only MiniMax M3 measured
Neither model has a result on the other 5 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index03
Math benchmarks for MiniMax M2.5 and MiniMax M3
BenchmarkMiniMax M2.5MiniMax M3Leader
AIME88.8%Not measuredOnly MiniMax M2.5 measured
OTIS Mock AIMENot measured71.1%Only MiniMax M3 measured
ProofBench4.0%18.0%MiniMax M3
LMArena Text: math13781429MiniMax M3
LMArena Text: mathematical work13781448MiniMax M3
Neither model has a result on the other 8 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index08
Knowledge benchmarks for MiniMax M2.5 and MiniMax M3
BenchmarkMiniMax M2.5MiniMax M3Leader
MMLU Pro80.1%84.2%MiniMax M3
LegalBench (Vals AI)80.0%85.4%MiniMax M3
CaseLaw (Vals AI)53.5%Not measuredOnly MiniMax M2.5 measured
CorpFin (Vals AI)59.6%68.1%MiniMax M3
TaxEval (Vals AI)68.2%72.7%MiniMax M3
MortgageTax (Vals AI)Not measured68.4%Only MiniMax M3 measured
MedQA (Vals AI)92.5%Not measuredOnly MiniMax M2.5 measured
MedCode (Vals AI)Not measured46.3%Only MiniMax M3 measured
LMArena Text: science13691452MiniMax M3
LMArena Text: medicine and healthcare13571441MiniMax M3
LMArena Text: legal and government13661437MiniMax M3
LMArena Text: business and finance13561434MiniMax M3
Neither model has a result on the other 5 boards here
Web designLike-for-likePublic boards, shown as published, kept out of the index07
Web design benchmarks for MiniMax M2.5 and MiniMax M3
BenchmarkMiniMax M2.5MiniMax M3Leader
Design Arena: websites12221257MiniMax M3
Design Arena: UI components11721246MiniMax M3
Design Arena: game dev11761210MiniMax M3
Design Arena: data viz11711235MiniMax M3
Design Arena: 3D11701209MiniMax M3
Design Arena: all code categories12071247MiniMax M3
Design Arena: SVG11471161MiniMax M3
Design Arena: ASCII artNot measured1161Only MiniMax M3 measured
LiveBench designNot measured66.2%Only MiniMax M3 measured
GeneralLike-for-likePublic boards, shown as published, kept out of the index019
General benchmarks for MiniMax M2.5 and MiniMax M3
BenchmarkMiniMax M2.5MiniMax M3Leader
LiveBench data analysisNot measured76.2%Only MiniMax M3 measured
Public Benefits Bench (Vals AI)Not measured64.1%Only MiniMax M3 measured
SAGE (Vals AI)Not measured50.6%Only MiniMax M3 measured
MedScribe (Vals AI)Not measured87.3%Only MiniMax M3 measured
Vals Multimodal IndexNot measured60.0%Only MiniMax M3 measured
IFBench71.6%82.9%MiniMax M3
Long context reasoning (AA-LCR)73.3%83.0%MiniMax M3
DTBenchNot measured78.9%Only MiniMax M3 measured
LMCANot measured33.7%Only MiniMax M3 measured
CL-bench11.4%Not measuredOnly MiniMax M2.5 measured
Surface Evolver BenchNot measured55.0%Only MiniMax M3 measured
CL-bench Life6.3%Not measuredOnly MiniMax M2.5 measured
LMArena Text13591433MiniMax M3
LMArena Text: instruction following13531433MiniMax M3
LMArena Text: creative writing13311404MiniMax M3
LMArena VisionNot measured1251Only MiniMax M3 measured
LMArena Text: English13801443MiniMax M3
LMArena Text: non-English13381420MiniMax M3
LMArena Text: Chinese13931466MiniMax M3
LMArena Text: German13611428MiniMax M3
LMArena Text: Japanese11711380MiniMax M3
LMArena Text: Korean12331373MiniMax M3
LMArena Text: French13631444MiniMax M3
LMArena Text: Spanish13531433MiniMax M3
LMArena Text: Russian13591426MiniMax M3
LMArena Text: Polish13341423MiniMax M3
LMArena Text: multi-turn13641443MiniMax M3
LMArena Text: longer queries13661446MiniMax M3
LMArena Text: writing and language13451419MiniMax M3
LMArena Text: entertainment and media13291403MiniMax M3
LMArena Vision: EnglishNot measured1263Only MiniMax M3 measured
LMArena Vision: ChineseNot measured1311Only MiniMax M3 measured
LMArena Vision: reading textNot measured1259Only MiniMax M3 measured
LMArena Vision: humorNot measured1235Only MiniMax M3 measured
LMArena Vision: recognizing thingsNot measured1197Only MiniMax M3 measured
LMArena Vision: creative writingNot measured1234Only MiniMax M3 measured
LMArena DocumentNot measured1435Only MiniMax M3 measured
Neither model has a result on the other 9 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 M3
BenchmarkMiniMax M2.5MiniMax M3Leader
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 M3$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 M3$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 M3$0.072Fits in one request

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 M3
SpecificationMiniMax M2.5MiniMax M3
MakerMiniMaxMiniMax
WeightsOpen weightsOpen weights
ReleasedFebruary 12, 2026May 31, 2026
Context window205K1.05M
Input $/Mtok$0.27$0.30
Output $/Mtok$0.95$1.20
Coding index59.457.8
Independent measurements behind it22
Cost per solved task$0.62$0.56
SourceOpenRouter listingOpenRouter listing

Questions

Is MiniMax M2.5 better than MiniMax M3 for coding?

MiniMax M2.5 and MiniMax M3 are level for coding, 59.4 against 57.8 on the Vibe Coding Index, so price and context decide. Pick MiniMax M3 when a solved task costs $0.56 against $0.62, or when the context window is larger, 1.05M against 205K tokens. Pick MiniMax M2.5 when output tokens cost 1.3x less, or when the benchmark that matters is τ²-Bench Telecom, where it leads.

Which is cheaper, MiniMax M2.5 or MiniMax M3?

MiniMax M2.5 costs $0.27 per million input tokens and $0.95 per million output tokens; MiniMax M3 costs $0.30 and $1.20. Per solved SWE-bench Verified task, MiniMax M2.5 costs $0.62 and MiniMax M3 $0.56.

Which has the bigger context window, MiniMax M2.5 or MiniMax M3?

MiniMax M3, with 1.05M tokens against 205K.

How many benchmarks is the MiniMax M2.5 vs MiniMax M3 comparison based on?

57 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 M3's on 2. Where only one model has a result, the page says so instead of naming a winner.

Compare next

Embed this comparison on your site

Free under CC BY 4.0, while the source link stays visible. Preview the frame · Terms and other embeds

<iframe src="https://llmbenchmark.ai/embed/compare/minimax-minimax-m2-5-vs-minimax-minimax-m3" title="MiniMax M2.5 vs MiniMax M3, LLMBenchmark.ai" width="100%" height="360" style="border:0;max-width:100%" loading="lazy"></iframe>
<p style="font:13px/1.4 sans-serif;margin:6px 0 0">Source: <a href="https://llmbenchmark.ai/compare/minimax-minimax-m2-5-vs-minimax-minimax-m3">LLMBenchmark.ai, MiniMax M2.5 vs MiniMax M3</a></p>