MiniMax M2.7 vs Hy3 preview
Updated October 21, 2026. 36 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick MiniMax M2.7 for coding: it scores 28.4 on the Vibe Coding Index against 22.4 for Hy3 preview.
Pick Hy3 preview when output tokens cost 2.0x less, or when the context window is larger, 262K against 205K tokens.
- Shared results
- 36
- MiniMax M2.7 only
- 38
- Hy3 preview only
- 0
- Like-for-like categories
- 7 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.
Coding
Vibe Coding Index, four public coding benchmarks
MiniMax M2.7
MiniMax M2.7 scores 28.4 against 22.4.
One benchmark behind each score
Intelligence
AA Intelligence Index
Hy3 preview
Hy3 preview leads, 25.3 against 22.8.
Same board, both measured
Agentic work
AA Agentic Index
Hy3 preview
Hy3 preview leads, 24.1 against 15.3.
Same board, both measured
Reasoning
GPQA Diamond
No call
Only MiniMax M2.7 has a GPQA Diamond result, 86.6%.
Directional only
Math
AIME
No call
Only MiniMax M2.7 has a AIME result, 91.0%.
Directional only
Knowledge
MMLU Pro
No call
Only MiniMax M2.7 has a MMLU Pro result, 80.4%.
Directional only
Web design
Design Arena: websites
No call
Only MiniMax M2.7 has a Design Arena: websites result, 1243.
Directional only
Long documents
Listed context window
Hy3 preview
Hy3 preview takes 262K tokens against 205K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Hy3 preview
MiniMax M2.7 $0.0009, Hy3 preview $0.0005 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only MiniMax M2.7 has the number, $0.64.
Directional only
At a glance
- Coding index
- MiniMax M2.7 28.4 from 1 benchmark, Hy3 preview 22.4 from 1 benchmark.
- Widest category
- Agentic work, 8.3 points apart on average over 2 shared benchmarks.
- Widest single gap
- Long context reasoning (AA-LCR): MiniMax M2.7 78.3% against Hy3 preview 64.7%, 13.7 points apart on a 0 to 100 scale.
- Output price
- MiniMax M2.7 $1.20 against Hy3 preview $0.60 per million output tokens, 2.0x apart.
- Context window
- MiniMax M2.7 takes 205K tokens, Hy3 preview 262K.
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.
| Category | MiniMax M2.7 | Bars | Hy3 preview | Ties |
|---|---|---|---|---|
| Coding | 6 | 1 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 0 | 2 | ||
| Reasoning | 5 | 0 | ||
| Math | 0 | 2 | ||
| Knowledge | 3 | 1 | ||
| General | 12 | 3 | ||
| All shared boards | 26 | 10 |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index61
| Benchmark | MiniMax M2.7 | Hy3 preview | Leader |
|---|---|---|---|
| LMArena WebDev | 1399 | 1357 | MiniMax M2.7 |
| AA Coding Index | 52.6 | 58.8 | Hy3 preview |
| LiveCodeBench (Vals AI) | 79.9% | Not measured | Only MiniMax M2.7 measured |
| SWE-bench Verified (Vals AI) | 73.8% | Not measured | Only MiniMax M2.7 measured |
| Vibe Code Bench (Vals AI) | 11.9% | Not measured | Only MiniMax M2.7 measured |
| Code migration (Vals AI) | 8.7% | Not measured | Only MiniMax M2.7 measured |
| ProgramBench (Vals AI) | 0.0% | Not measured | Only MiniMax M2.7 measured |
| Terminal-Bench 2.1 (Vals AI) | 48.7% | Not measured | Only MiniMax M2.7 measured |
| SciCode | 50.1% | Not measured | Only MiniMax M2.7 measured |
| Terminal-Bench Hard (Artificial Analysis run) | 39.4% | 34.1% | MiniMax M2.7 |
| Terminal-Bench | 45.1% | Not measured | Only MiniMax M2.7 measured |
| WeirdML | 37.0% | Not measured | Only MiniMax M2.7 measured |
| ALE-Bench | 599.3 | Not measured | Only MiniMax M2.7 measured |
| LMArena Text: coding | 1454 | 1421 | MiniMax M2.7 |
| LMArena Text: software and IT | 1446 | 1423 | MiniMax M2.7 |
| LMArena WebDev: React | 1388 | 1349 | MiniMax M2.7 |
| LMArena WebDev: HTML | 1401 | 1356 | MiniMax M2.7 |
| Neither model has a result on the other 21 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | MiniMax M2.7 | Hy3 preview | Leader |
|---|---|---|---|
| AA Intelligence Index | 22.8 | 25.3 | Hy3 preview |
| Epoch Capabilities Index | 145.8 | Not measured | Only MiniMax M2.7 measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | MiniMax M2.7 | Hy3 preview | Leader |
|---|---|---|---|
| AA Agentic Index | 15.3 | 24.1 | Hy3 preview |
| Finance Agent (Vals AI) | 27.9% | Not measured | Only MiniMax M2.7 measured |
| Tax Agent Bench (Vals AI) | 6.9% | Not measured | Only MiniMax M2.7 measured |
| Legal Research Bench (Vals AI) | 10.6% | Not measured | Only MiniMax M2.7 measured |
| Excel Modeling Benchmark (Vals AI) | 28.4% | Not measured | Only MiniMax M2.7 measured |
| τ²-Bench Telecom | 84.8% | 92.7% | Hy3 preview |
| Neither model has a result on the other 33 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index50
| Benchmark | MiniMax M2.7 | Hy3 preview | Leader |
|---|---|---|---|
| GPQA Diamond | 86.6% | Not measured | Only MiniMax M2.7 measured |
| GPQA Diamond (Artificial Analysis run) | 87.4% | 86.7% | MiniMax M2.7 |
| Humanity's Last Exam (Artificial Analysis run) | 29.6% | 27.8% | MiniMax M2.7 |
| LMArena Text: hard prompts | 1422 | 1413 | MiniMax M2.7 |
| LMArena Text: expert | 1443 | 1422 | MiniMax M2.7 |
| LMArena Text: hard prompts in English | 1441 | 1416 | MiniMax M2.7 |
| Neither model has a result on the other 15 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | MiniMax M2.7 | Hy3 preview | Leader |
|---|---|---|---|
| AIME | 91.0% | Not measured | Only MiniMax M2.7 measured |
| ProofBench | 3.0% | Not measured | Only MiniMax M2.7 measured |
| LMArena Text: math | 1421 | 1424 | Hy3 preview |
| LMArena Text: mathematical work | 1437 | 1442 | Hy3 preview |
| Neither model has a result on the other 9 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index31
| Benchmark | MiniMax M2.7 | Hy3 preview | Leader |
|---|---|---|---|
| MMLU Pro | 80.4% | Not measured | Only MiniMax M2.7 measured |
| LegalBench (Vals AI) | 84.0% | Not measured | Only MiniMax M2.7 measured |
| CaseLaw (Vals AI) | 60.9% | Not measured | Only MiniMax M2.7 measured |
| CorpFin (Vals AI) | 61.2% | Not measured | Only MiniMax M2.7 measured |
| TaxEval (Vals AI) | 66.6% | Not measured | Only MiniMax M2.7 measured |
| MedCode (Vals AI) | 34.4% | Not measured | Only MiniMax M2.7 measured |
| LMArena Text: science | 1416 | 1416 | MiniMax M2.7 |
| LMArena Text: medicine and healthcare | 1409 | 1405 | MiniMax M2.7 |
| LMArena Text: legal and government | 1403 | 1430 | Hy3 preview |
| LMArena Text: business and finance | 1407 | 1397 | MiniMax M2.7 |
| Neither model has a result on the other 7 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | MiniMax M2.7 | Hy3 preview | Leader |
|---|---|---|---|
| Design Arena: websites | 1243 | Not measured | Only MiniMax M2.7 measured |
| Design Arena: UI components | 1210 | Not measured | Only MiniMax M2.7 measured |
| Design Arena: game dev | 1210 | Not measured | Only MiniMax M2.7 measured |
| Design Arena: data viz | 1233 | Not measured | Only MiniMax M2.7 measured |
| Design Arena: 3D | 1195 | Not measured | Only MiniMax M2.7 measured |
| Design Arena: all code categories | 1233 | Not measured | Only MiniMax M2.7 measured |
| Design Arena: SVG | 1136 | Not measured | Only MiniMax M2.7 measured |
| Design Arena: ASCII art | 1136 | Not measured | Only 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 index123
| Benchmark | MiniMax M2.7 | Hy3 preview | Leader |
|---|---|---|---|
| MedScribe (Vals AI) | 79.9% | Not measured | Only MiniMax M2.7 measured |
| IFBench | 75.7% | 63.1% | MiniMax M2.7 |
| Long context reasoning (AA-LCR) | 78.3% | 64.7% | MiniMax M2.7 |
| ExploitBench | 13.3% | Not measured | Only MiniMax M2.7 measured |
| LMArena Text | 1405 | 1401 | MiniMax M2.7 |
| LMArena Text: instruction following | 1405 | 1383 | MiniMax M2.7 |
| LMArena Text: creative writing | 1354 | 1339 | MiniMax M2.7 |
| LMArena Text: English | 1426 | 1409 | MiniMax M2.7 |
| LMArena Text: non-English | 1382 | 1390 | Hy3 preview |
| LMArena Text: Chinese | 1442 | 1435 | MiniMax M2.7 |
| LMArena Text: German | 1398 | Not measured | Only MiniMax M2.7 measured |
| LMArena Text: Japanese | 1260 | Not measured | Only MiniMax M2.7 measured |
| LMArena Text: Korean | 1314 | Not measured | Only MiniMax M2.7 measured |
| LMArena Text: French | 1424 | 1425 | Hy3 preview |
| LMArena Text: Spanish | 1402 | 1399 | MiniMax M2.7 |
| LMArena Text: Russian | 1383 | 1399 | Hy3 preview |
| LMArena Text: Polish | 1387 | Not measured | Only MiniMax M2.7 measured |
| LMArena Text: multi-turn | 1412 | 1411 | MiniMax M2.7 |
| LMArena Text: longer queries | 1418 | 1402 | MiniMax M2.7 |
| LMArena Text: writing and language | 1376 | 1370 | MiniMax M2.7 |
| LMArena Text: entertainment and media | 1360 | 1352 | MiniMax M2.7 |
| Neither model has a result on the other 25 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | MiniMax M2.7 | Hy3 preview | Leader |
|---|---|---|---|
| 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
Hy3 preview costs less
Repository review
50K input + 3K output tokens
Hy3 preview costs less
Agent loop
200K input + 10K output tokens
Hy3 preview 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.
| Specification | MiniMax M2.7 | Tencent Hy3 preview |
|---|---|---|
| Maker | MiniMax | Tencent |
| Weights | Open weights | Open weights |
| Released | March 18, 2026 | April 22, 2026 |
| Context window | 205K | 262K |
| Input $/Mtok | $0.30 | $0.18 |
| Output $/Mtok | $1.20 | $0.60 |
| Coding index | 28.4 | 22.4 |
| Independent measurements behind it | 1 | 1 |
| Cost per solved task | $0.64 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is MiniMax M2.7 better than Hy3 preview for coding?
Pick MiniMax M2.7 for coding: it scores 28.4 on the Vibe Coding Index against 22.4 for Hy3 preview. Pick Hy3 preview when output tokens cost 2.0x less, or when the context window is larger, 262K against 205K tokens.
Which is cheaper, MiniMax M2.7 or Hy3 preview?
MiniMax M2.7 costs $0.30 per million input tokens and $1.20 per million output tokens; Hy3 preview costs $0.18 and $0.60. Cost per solved task is known for MiniMax M2.7 only.
Which has the bigger context window, MiniMax M2.7 or Hy3 preview?
Hy3 preview, with 262K tokens against 205K.
How many benchmarks is the MiniMax M2.7 vs Hy3 preview comparison based on?
36 benchmarks have a result for both models as of October 21, 2026. MiniMax M2.7's coding index rests on 1 independent benchmark and Hy3 preview's on 1. Where only one model has a result, the page says so instead of naming a winner.