Enter opens the highlighted result. Esc closes.

Phi 4 vs MiniMax M2.5

MiniMax M2.5 leads Phi 4 on the Artificial Analysis Intelligence Index, 22.8 to 5.9, and Phi 4 costs 8.6x less per output token ($0.14 vs $1.20 per million). Updated October 21, 2026; 35 benchmarks have a result for both.

Replace Phi 4 Replace MiniMax M2.5
Phi 4Microsoft
Coding
No data
Overall
5.9
Output $/M
$0.14
0
MiniMax M2.5MiniMax
Coding
37.2
Overall
22.8
Output $/M
$1.20
35
vs boards won of 35 shared

Phi 4 or MiniMax M2.5: which to pick

Only MiniMax M2.5 has a coding score, 37.2 on the Vibe Coding Index; Phi 4 has none, so the two cannot be ranked on coding.

Pick Phi 4 for output tokens 8.6x cheaper ($0.14 vs $1.20 per million).

Model APhi 4MicrosoftNot benchmarkedNo coding benchmarkNo cost per solved taskModel BMiniMax M2.5MiniMax37.2/1003 independent benchmarks$0.62 per solved task
Shared results
35
Phi 4 only
10
MiniMax M2.5 only
32
Like-for-like categories
7 of 9

Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.

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
Phi 4 not benchmarked, MiniMax M2.5 37.2 from 3 benchmarks.
Widest category
Agentic work, 95.3 points apart on average over 1 shared benchmark.
Widest single gap
τ²-Bench Telecom: Phi 4 0.0% against MiniMax M2.5 95.3%, 95.3 points apart on a 0 to 100 scale.
Output price
Phi 4 $0.14 against MiniMax M2.5 $1.20 per million output tokens, 8.6x apart.
Context window
Phi 4 takes 16K tokens, MiniMax M2.5 205K.

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
CategoryPhi 4BarsMiniMax M2.5Ties
Coding03
Intelligence02
Agentic work01
Reasoning05
Math02
Knowledge04
General018
All shared boards035
CodingLike-for-likeSix of these boards feed the Vibe Coding Index03
Coding benchmarks for Phi 4 and MiniMax M2.5
BenchmarkPhi 4MiniMax M2.5Leader
SWE-bench VerifiedNot measured75.8%Only MiniMax M2.5 measured
SWE-bench MultilingualNot measured68.3%Only MiniMax M2.5 measured
LMArena WebDevNot measured1386Only MiniMax M2.5 measured
LiveCodeBench (Vals AI)Not measured79.2%Only MiniMax M2.5 measured
SWE-bench Verified (Vals AI)Not measured74.2%Only MiniMax M2.5 measured
Vibe Code Bench (Vals AI)Not measured14.8%Only MiniMax M2.5 measured
LiveCodeBench (Artificial Analysis run)23.1%Not measuredOnly Phi 4 measured
Terminal-Bench Hard (Artificial Analysis run)3.8%34.8%MiniMax M2.5
Terminal-BenchNot measured42.7%Only MiniMax M2.5 measured
ALE-BenchNot measured618.2Only MiniMax M2.5 measured
LMArena Text: coding12311380MiniMax M2.5
LMArena Text: software and IT12281383MiniMax M2.5
LMArena WebDev: ReactNot measured1374Only MiniMax M2.5 measured
LMArena WebDev: HTMLNot measured1414Only MiniMax M2.5 measured
Neither model has a result on the other 24 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
Intelligence benchmarks for Phi 4 and MiniMax M2.5
BenchmarkPhi 4MiniMax M2.5Leader
AA Intelligence Index5.922.8MiniMax M2.5
Epoch Capabilities Index130.5146.6MiniMax M2.5
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
Agentic work benchmarks for Phi 4 and MiniMax M2.5
BenchmarkPhi 4MiniMax M2.5Leader
τ²-Bench Telecom0.0%95.3%MiniMax M2.5
BALROG11.6%Not measuredOnly Phi 4 measured
Vending-Bench 2Not measured-$23Only MiniMax M2.5 measured
Neither model has a result on the other 36 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index05
Reasoning benchmarks for Phi 4 and MiniMax M2.5
BenchmarkPhi 4MiniMax M2.5Leader
GPQA DiamondNot measured82.1%Only MiniMax M2.5 measured
GPQA Diamond (Artificial Analysis run)57.5%84.8%MiniMax M2.5
Humanity's Last Exam (Artificial Analysis run)3.8%20.5%MiniMax M2.5
GPQA Diamond (Epoch AI run)56.1%Not measuredOnly Phi 4 measured
ARC-AGINot measured63.7%Only MiniMax M2.5 measured
ARC-AGI-2Not measured4.9%Only MiniMax M2.5 measured
LMArena Text: hard prompts12201372MiniMax M2.5
LMArena Text: expert12031378MiniMax M2.5
LMArena Text: hard prompts in English12301384MiniMax M2.5
Neither model has a result on the other 12 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index02
Math benchmarks for Phi 4 and MiniMax M2.5
BenchmarkPhi 4MiniMax M2.5Leader
AIMENot measured88.8%Only MiniMax M2.5 measured
AA Math Index18.0Not measuredOnly Phi 4 measured
AIME 2025 (Artificial Analysis run)18.0%Not measuredOnly Phi 4 measured
OTIS Mock AIME13.8%Not measuredOnly Phi 4 measured
ProofBenchNot measured4.0%Only MiniMax M2.5 measured
MATH level 564.9%Not measuredOnly Phi 4 measured
LMArena Text: math12461378MiniMax M2.5
LMArena Text: mathematical work12541378MiniMax M2.5
Neither model has a result on the other 5 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index04
Knowledge benchmarks for Phi 4 and MiniMax M2.5
BenchmarkPhi 4MiniMax M2.5Leader
MMLU ProNot measured80.1%Only MiniMax M2.5 measured
LegalBench (Vals AI)Not measured80.0%Only MiniMax M2.5 measured
CaseLaw (Vals AI)Not measured53.5%Only MiniMax M2.5 measured
CorpFin (Vals AI)Not measured59.6%Only MiniMax M2.5 measured
TaxEval (Vals AI)Not measured68.2%Only MiniMax M2.5 measured
MedQA (Vals AI)Not measured92.5%Only MiniMax M2.5 measured
MMLU Pro (Artificial Analysis run)71.4%Not measuredOnly Phi 4 measured
MMLU84.8%Not measuredOnly Phi 4 measured
LMArena Text: science12061369MiniMax M2.5
LMArena Text: medicine and healthcare12021357MiniMax M2.5
LMArena Text: legal and government12411366MiniMax M2.5
LMArena Text: business and finance11991356MiniMax M2.5
Neither model has a result on the other 5 boards here
Web designDirectional onlyPublic boards, shown as published, kept out of the index7 boards, none shared
Web design benchmarks for Phi 4 and MiniMax M2.5
BenchmarkPhi 4MiniMax M2.5Leader
Design Arena: websitesNot measured1222Only MiniMax M2.5 measured
Design Arena: UI componentsNot measured1173Only MiniMax M2.5 measured
Design Arena: game devNot measured1176Only MiniMax M2.5 measured
Design Arena: data vizNot measured1171Only MiniMax M2.5 measured
Design Arena: 3DNot measured1170Only MiniMax M2.5 measured
Design Arena: all code categoriesNot measured1207Only MiniMax M2.5 measured
Design Arena: SVGNot measured1149Only MiniMax M2.5 measured
Neither model has a result on the other 2 boards here
GeneralLike-for-likePublic boards, shown as published, kept out of the index018
General benchmarks for Phi 4 and MiniMax M2.5
BenchmarkPhi 4MiniMax M2.5Leader
IFBench23.5%71.6%MiniMax M2.5
Long context reasoning (AA-LCR)0.0%73.3%MiniMax M2.5
Creative writing (Lech Mazur)62.6%Not measuredOnly Phi 4 measured
CL-benchNot measured11.4%Only MiniMax M2.5 measured
CL-bench LifeNot measured6.3%Only MiniMax M2.5 measured
LMArena Text12171359MiniMax M2.5
LMArena Text: instruction following12011353MiniMax M2.5
LMArena Text: creative writing11811331MiniMax M2.5
LMArena Text: English12311380MiniMax M2.5
LMArena Text: non-English11971338MiniMax M2.5
LMArena Text: Chinese12121393MiniMax M2.5
LMArena Text: German12221361MiniMax M2.5
LMArena Text: Japanese11581171MiniMax M2.5
LMArena Text: Korean11511233MiniMax M2.5
LMArena Text: French12241363MiniMax M2.5
LMArena Text: Spanish12341353MiniMax M2.5
LMArena Text: Russian12091359MiniMax M2.5
LMArena Text: PolishNot measured1334Only MiniMax M2.5 measured
LMArena Text: multi-turn12061364MiniMax M2.5
LMArena Text: longer queries12171366MiniMax M2.5
LMArena Text: writing and language11981345MiniMax M2.5
LMArena Text: entertainment and media11621329MiniMax M2.5
Neither model has a result on the other 24 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for Phi 4 and MiniMax M2.5
BenchmarkPhi 4MiniMax M2.5Leader
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

Phi 4$0.0001Fits in one request
MiniMax M2.5$0.0009Fits in one request

Phi 4 costs less

Repository review

50K input + 3K output tokens

Phi 4$0.0039Exceeds the context window
MiniMax M2.5$0.019Fits in one request

Phi 4 costs less

Agent loop

200K input + 10K output tokens

Phi 4$0.015Exceeds the context window
MiniMax M2.5$0.072Exceeds the context window

Phi 4 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 Phi 4 and MiniMax M2.5
SpecificationMicrosoft Phi 4MiniMax M2.5
MakerMicrosoftMiniMax
WeightsOpen weightsOpen weights
ReleasedJanuary 10, 2025February 12, 2026
Context window16K205K
Input $/Mtok$0.07$0.30
Output $/Mtok$0.14$1.20
Coding indexNot benchmarked37.2
Independent measurements behind it03
Cost per solved taskNot computable$0.62
SourceOpenRouter listingOpenRouter listing

Questions

Is Phi 4 better than MiniMax M2.5 for coding?

Only MiniMax M2.5 has a coding score, 37.2 on the Vibe Coding Index; Phi 4 has none, so the two cannot be ranked on coding. Pick Phi 4 for output tokens 8.6x cheaper ($0.14 vs $1.20 per million).

Which is cheaper, Phi 4 or MiniMax M2.5?

Phi 4 costs $0.07 per million input tokens and $0.14 per million output tokens; MiniMax M2.5 costs $0.30 and $1.20. Cost per solved task is known for MiniMax M2.5 only.

Which has the bigger context window, Phi 4 or MiniMax M2.5?

MiniMax M2.5, with 205K tokens against 16K.

How many benchmarks is the Phi 4 vs MiniMax M2.5 comparison based on?

35 benchmarks have a result for both models as of October 21, 2026. Phi 4's coding index rests on 0 independent benchmarks and MiniMax M2.5's on 3. Where only one model has a result, the page says so instead of naming a winner.

Compare next