Enter opens the highlighted result. Esc closes.

MiniMax M3 vs Hy4 flash

Hy4 flash leads MiniMax M3 on coding, 44.5 to 33.6 on the Vibe Coding Index, and Hy4 flash costs 8.5x less per output token ($0.1405 vs $1.20 per million). Updated November 22, 2026; 10 benchmarks have a result for both.

MiniMax M3MiniMax
Coding
33.6
Overall
52.7
Output $/M
$1.20
2
Hy4 flashTencent
Coding
44.5
Overall
No data
Output $/M
$0.1405
8
vs boards won of 10 shared

MiniMax M3 or Hy4 flash: which to pick

Pick Hy4 flash for coding: it scores 44.5 on the Vibe Coding Index against 33.6 for MiniMax M3, though on 2 benchmarks against 5.

Pick MiniMax M3 for work measured by Design Arena: websites or Design Arena: all code categories, where it leads.

Shared results
10
MiniMax M3 only
106
Hy4 flash only
2
Like-for-like categories
3 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
MiniMax M3 33.6 from 5 benchmarks, Hy4 flash 44.5 from 2 benchmarks.
Widest category
Coding, 22.6 points apart on average over 3 shared benchmarks.
Widest single gap
Vibe Code Bench (Vals AI): MiniMax M3 47.6% against Hy4 flash 74.8%, 27.3 points apart on a 0 to 100 scale.
Output price
MiniMax M3 $1.20 against Hy4 flash $0.1405 per million output tokens, 8.5x apart.
Context window
Both take 1.05M 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 M3BarsHy4 flashTies
Coding03
Agentic work02
Web design23
All shared boards28
CodingLike-for-likeSix of these boards feed the Vibe Coding Index03
Coding benchmarks for MiniMax M3 and Hy4 flash
BenchmarkMiniMax M3Hy4 flashLeader
LMArena WebDev1484Not measuredOnly MiniMax M3 measured
LiveCodeBench (Vals AI)82.2%Not measuredOnly MiniMax M3 measured
SWE-bench Verified (Vals AI)75.0%Not measuredOnly MiniMax M3 measured
Terminal-Bench 4 (Vals AI)1.0%15.2%Hy4 flash
Vibe Code Bench (Vals AI)47.6%74.8%Hy4 flash
Code migration (Vals AI)19.9%Not measuredOnly MiniMax M3 measured
Terminal-Bench 2.1 (Vals AI)53.6%Not measuredOnly MiniMax M3 measured
SWE-bench Multimodal (Vals AI)33.3%Not measuredOnly MiniMax M3 measured
Vibe Code Bench 1-100 (Vals AI)9.2%Not measuredOnly MiniMax M3 measured
SkillsBench (Vals AI)51.5%Not measuredOnly MiniMax M3 measured
CyberBench (Vals AI)65.0%Not measuredOnly MiniMax M3 measured
SciCode47.1%Not measuredOnly MiniMax M3 measured
Terminal-Bench Hard (Artificial Analysis run)42.4%Not measuredOnly MiniMax M3 measured
Terminal-Bench 4.0 (Artificial Analysis run)2.0%Not measuredOnly MiniMax M3 measured
FrontierCode 2.021.9%Not measuredOnly MiniMax M3 measured
FrontierCode14.7%Not measuredOnly MiniMax M3 measured
ALE-Bench640.0Not measuredOnly MiniMax M3 measured
LMArena Text: coding1467Not measuredOnly MiniMax M3 measured
LMArena Text: software and IT1463Not measuredOnly MiniMax M3 measured
LMArena WebDev: React1479Not measuredOnly MiniMax M3 measured
LMArena WebDev: HTML1461Not measuredOnly MiniMax M3 measured
LMArena WebDev: from an image1474Not measuredOnly MiniMax M3 measured
GameBench 228.850.5Hy4 flash
Terminal-Bench 4.07.3%Not measuredOnly MiniMax M3 measured
Neither model has a result on the other 28 boards here
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
Intelligence benchmarks for MiniMax M3 and Hy4 flash
BenchmarkMiniMax M3Hy4 flashLeader
AA Intelligence Index18.9Not measuredOnly MiniMax M3 measured
Epoch Capabilities Index147.8Not measuredOnly MiniMax M3 measured
Agentic workLike-for-likePublic boards, shown as published, kept out of the index02
Agentic work benchmarks for MiniMax M3 and Hy4 flash
BenchmarkMiniMax M3Hy4 flashLeader
Design Arena: agents building web apps1179Not measuredOnly MiniMax M3 measured
Design Arena: agents building full-stack apps1179Not measuredOnly MiniMax M3 measured
Design Arena: agents building mobile apps1172Not measuredOnly MiniMax M3 measured
Design Arena: agents building Android apps1155Not measuredOnly MiniMax M3 measured
Design Arena: agents building games1113Not measuredOnly MiniMax M3 measured
Design Arena: HTML slides1084Not measuredOnly MiniMax M3 measured
Design Arena: PowerPoint slides1167Not measuredOnly MiniMax M3 measured
Design Arena: spreadsheets1168Not measuredOnly MiniMax M3 measured
Finance Agent (Vals AI)48.3%Not measuredOnly MiniMax M3 measured
Tax Agent Bench (Vals AI)20.6%23.3%Hy4 flash
Legal Research Bench (Vals AI)29.8%38.5%Hy4 flash
Harvey's Legal Agent Benchmark (Vals AI)Not measured5.4%Only Hy4 flash measured
Excel Modeling Benchmark (Vals AI)47.8%Not measuredOnly MiniMax M3 measured
τ²-Bench Telecom88.9%Not measuredOnly MiniMax M3 measured
APEX-Agents37.7%Not measuredOnly MiniMax M3 measured
OSWorld 2.04.6%Not measuredOnly MiniMax M3 measured
Vending-Bench 2$2,158Not measuredOnly MiniMax M3 measured
Neither model has a result on the other 27 boards here
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index15 boards, none shared
Reasoning benchmarks for MiniMax M3 and Hy4 flash
BenchmarkMiniMax M3Hy4 flashLeader
GPQA Diamond92.7%Not measuredOnly MiniMax M3 measured
MMMU Pro (Vals AI)81.2%Not measuredOnly MiniMax M3 measured
GPQA Diamond (Artificial Analysis run)92.9%Not measuredOnly MiniMax M3 measured
Humanity's Last Exam (Artificial Analysis run)39.0%Not measuredOnly MiniMax M3 measured
HLE-Diamond (Artificial Analysis run)16.0%Not measuredOnly MiniMax M3 measured
CritPt v2, hard subset (Artificial Analysis run)28.0%Not measuredOnly MiniMax M3 measured
GPQA Diamond (Epoch AI run)90.9%Not measuredOnly MiniMax M3 measured
SimpleBench45.8%Not measuredOnly MiniMax M3 measured
Chess puzzles17.0%Not measuredOnly MiniMax M3 measured
Mystery game puzzles8.0%Not measuredOnly MiniMax M3 measured
LMArena Text: hard prompts1441Not measuredOnly MiniMax M3 measured
LMArena Text: expert1460Not measuredOnly MiniMax M3 measured
LMArena Text: hard prompts in English1452Not measuredOnly MiniMax M3 measured
LMArena Vision: diagrams1265Not measuredOnly MiniMax M3 measured
LMArena Vision: homework1266Not measuredOnly MiniMax M3 measured
Neither model has a result on the other 11 boards here
MathDirectional onlyPublic boards, shown as published, kept out of the index4 boards, none shared
Math benchmarks for MiniMax M3 and Hy4 flash
BenchmarkMiniMax M3Hy4 flashLeader
OTIS Mock AIME71.1%Not measuredOnly MiniMax M3 measured
ProofBench18.0%Not measuredOnly MiniMax M3 measured
LMArena Text: math1422Not measuredOnly MiniMax M3 measured
LMArena Text: mathematical work1441Not measuredOnly MiniMax M3 measured
Neither model has a result on the other 10 boards here
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index11 boards, none shared
Knowledge benchmarks for MiniMax M3 and Hy4 flash
BenchmarkMiniMax M3Hy4 flashLeader
MMLU Pro84.2%Not measuredOnly MiniMax M3 measured
LegalBench (Vals AI)85.4%Not measuredOnly MiniMax M3 measured
CorpFin (Vals AI)68.1%Not measuredOnly MiniMax M3 measured
TaxEval (Vals AI)72.7%Not measuredOnly MiniMax M3 measured
MortgageTax (Vals AI)68.4%Not measuredOnly MiniMax M3 measured
MedCode (Vals AI)46.3%Not measuredOnly MiniMax M3 measured
AA-Omniscience-5.5Not measuredOnly MiniMax M3 measured
LMArena Text: science1446Not measuredOnly MiniMax M3 measured
LMArena Text: medicine and healthcare1432Not measuredOnly MiniMax M3 measured
LMArena Text: legal and government1427Not measuredOnly MiniMax M3 measured
LMArena Text: business and finance1429Not measuredOnly MiniMax M3 measured
Neither model has a result on the other 10 boards here
Web designLike-for-likePublic boards, shown as published, kept out of the index23
Web design benchmarks for MiniMax M3 and Hy4 flash
BenchmarkMiniMax M3Hy4 flashLeader
Design Arena: websites12441227MiniMax M3
Design Arena: UI components12281262Hy4 flash
Design Arena: game dev11811202Hy4 flash
Design Arena: data viz1226Not measuredOnly MiniMax M3 measured
Design Arena: 3D11761209Hy4 flash
Design Arena: all code categories12281225MiniMax M3
Design Arena: SVG1140Not measuredOnly MiniMax M3 measured
Design Arena: ASCII art1140Not measuredOnly MiniMax M3 measured
Neither model has a result on the other 1 board here
GeneralDirectional onlyPublic boards, shown as published, kept out of the index37 boards, none shared
General benchmarks for MiniMax M3 and Hy4 flash
BenchmarkMiniMax M3Hy4 flashLeader
Vals IndexNot measured50.8%Only Hy4 flash measured
Public Benefits Bench (Vals AI)64.1%Not measuredOnly MiniMax M3 measured
SAGE (Vals AI)85.5%Not measuredOnly MiniMax M3 measured
MedScribe (Vals AI)87.3%Not measuredOnly MiniMax M3 measured
Vals Multimodal Index60.0%Not measuredOnly MiniMax M3 measured
IFBench82.9%Not measuredOnly MiniMax M3 measured
Long context reasoning (AA-LCR)83.0%Not measuredOnly MiniMax M3 measured
GDPval-AA1250Not measuredOnly MiniMax M3 measured
DTBench78.9%Not measuredOnly MiniMax M3 measured
LMCA33.7%Not measuredOnly MiniMax M3 measured
Surface Evolver Bench55.0%Not measuredOnly MiniMax M3 measured
LMArena Text1427Not measuredOnly MiniMax M3 measured
LMArena Text: instruction following1425Not measuredOnly MiniMax M3 measured
LMArena Text: creative writing1391Not measuredOnly MiniMax M3 measured
LMArena Vision1248Not measuredOnly MiniMax M3 measured
LMArena Text: English1439Not measuredOnly MiniMax M3 measured
LMArena Text: non-English1412Not measuredOnly MiniMax M3 measured
LMArena Text: Chinese1464Not measuredOnly MiniMax M3 measured
LMArena Text: German1426Not measuredOnly MiniMax M3 measured
LMArena Text: Japanese1354Not measuredOnly MiniMax M3 measured
LMArena Text: Korean1362Not measuredOnly MiniMax M3 measured
LMArena Text: French1436Not measuredOnly MiniMax M3 measured
LMArena Text: Spanish1421Not measuredOnly MiniMax M3 measured
LMArena Text: Russian1419Not measuredOnly MiniMax M3 measured
LMArena Text: Polish1414Not measuredOnly MiniMax M3 measured
LMArena Text: multi-turn1437Not measuredOnly MiniMax M3 measured
LMArena Text: longer queries1438Not measuredOnly MiniMax M3 measured
LMArena Text: writing and language1407Not measuredOnly MiniMax M3 measured
LMArena Text: entertainment and media1394Not measuredOnly MiniMax M3 measured
LMArena Vision: English1260Not measuredOnly MiniMax M3 measured
LMArena Vision: Chinese1305Not measuredOnly MiniMax M3 measured
LMArena Vision: reading text1255Not measuredOnly MiniMax M3 measured
LMArena Vision: humor1233Not measuredOnly MiniMax M3 measured
LMArena Vision: recognizing things1194Not measuredOnly MiniMax M3 measured
LMArena Vision: creative writing1231Not measuredOnly MiniMax M3 measured
LMArena Document1437Not measuredOnly MiniMax M3 measured
EQ-Bench 41140Not measuredOnly MiniMax M3 measured
Neither model has a result on the other 12 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for MiniMax M3 and Hy4 flash
BenchmarkMiniMax M3Hy4 flashLeader
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 M3$0.0009Fits in one request
Hy4 flash$0.0001Fits in one request

Hy4 flash costs less

Repository review

50K input + 3K output tokens

MiniMax M3$0.019Fits in one request
Hy4 flash$0.0022Fits in one request

Hy4 flash costs less

Agent loop

200K input + 10K output tokens

MiniMax M3$0.072Fits in one request
Hy4 flash$0.0084Fits in one request

Hy4 flash 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 M3 and Hy4 flash
SpecificationMiniMax M3Tencent Hy4 flash
MakerMiniMaxTencent
WeightsOpen weightsOpen weights
ReleasedMay 31, 2026November 16, 2026
Context window1.05M1.05M
Input $/Mtok$0.30$0.035
Output $/Mtok$1.20$0.1405
Coding index33.644.5
Independent measurements behind it52
Cost per solved task$0.56Not computable
SourceOpenRouter listingOpenRouter listing

Questions

Is MiniMax M3 better than Hy4 flash for coding?

Pick Hy4 flash for coding: it scores 44.5 on the Vibe Coding Index against 33.6 for MiniMax M3, though on 2 benchmarks against 5. Pick MiniMax M3 for work measured by Design Arena: websites or Design Arena: all code categories, where it leads.

Which is cheaper, MiniMax M3 or Hy4 flash?

MiniMax M3 costs $0.30 per million input tokens and $1.20 per million output tokens; Hy4 flash costs $0.035 and $0.1405. Cost per solved task is known for MiniMax M3 only.

Which has the bigger context window, MiniMax M3 or Hy4 flash?

Neither: both take 1.05M tokens.

How many benchmarks is the MiniMax M3 vs Hy4 flash comparison based on?

10 benchmarks have a result for both models as of November 22, 2026. MiniMax M3's coding index rests on 5 independent benchmarks and Hy4 flash'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-m3-vs-tencent-hy4-flash" title="MiniMax M3 vs Hy4 flash, 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-m3-vs-tencent-hy4-flash">LLMBenchmark.ai, MiniMax M3 vs Hy4 flash</a></p>