Kimi K2 Thinking vs MiniMax M3
Updated October 21, 2026. 17 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick Kimi K2 Thinking for coding: it scores 63.4 on the Vibe Coding Index against 57.8 for MiniMax M3, though on 1 benchmark against 2.
Pick MiniMax M3 when output tokens cost 2.1x less, or when a solved task costs $0.56 against $0.61.
- Shared results
- 17
- Kimi K2 Thinking only
- 13
- MiniMax M3 only
- 99
- 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
Kimi K2 Thinking
Kimi K2 Thinking scores 63.4 against 57.8.
1 and 2 benchmarks behind the scores
Intelligence
AA Intelligence Index
MiniMax M3
MiniMax M3 leads, 29.2 against 22.0.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only MiniMax M3 has a AA Agentic Index result, 29.5.
Directional only
Reasoning
GPQA Diamond
MiniMax M3
MiniMax M3 leads, 92.7% against 78.5%.
Same board, both measured
Math
AIME
No call
Only Kimi K2 Thinking has a AIME result, 85.4%.
Directional only
Knowledge
MMLU Pro
MiniMax M3
MiniMax M3 leads, 84.2% against 81.1%.
Same board, both measured
Web design
Design Arena: websites
MiniMax M3
MiniMax M3 leads, 1257 against 1113.
Same board, both measured
Long documents
Listed context window
MiniMax M3
MiniMax M3 takes 1.05M tokens against 262K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
MiniMax M3
Kimi K2 Thinking $0.0019, MiniMax M3 $0.0009 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
MiniMax M3
Kimi K2 Thinking $0.61, MiniMax M3 $0.56.
Published attempt cost and pass rate
At a glance
- Coding index
- Kimi K2 Thinking 63.4 from 1 benchmark, MiniMax M3 57.8 from 2 benchmarks.
- Widest category
- Web design, 18.0 points apart on average over 1 shared benchmark.
- Widest single gap
- LiveCodeBench (Vals AI): Kimi K2 Thinking 63.1% against MiniMax M3 82.2%, 19.0 points apart on a 0 to 100 scale.
- Output price
- Kimi K2 Thinking $2.50 against MiniMax M3 $1.20 per million output tokens, 2.1x apart.
- Context window
- Kimi K2 Thinking takes 262K 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.
| Category | Kimi K2 Thinking | Bars | MiniMax M3 | Ties |
|---|---|---|---|---|
| Coding | 0 | 4 | ||
| Intelligence | 0 | 2 | ||
| Agentic work | 1 | 0 | ||
| Reasoning | 0 | 3 | ||
| Knowledge | 0 | 4 | ||
| Web design | 0 | 1 | ||
| General | 0 | 2 | ||
| All shared boards | 1 | 16 |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index04
| Benchmark | Kimi K2 Thinking | MiniMax M3 | Leader |
|---|---|---|---|
| SWE-bench Verified | 63.4% | Not measured | Only Kimi K2 Thinking measured |
| LiveBench coding | Not measured | 70.9% | Only MiniMax M3 measured |
| LMArena WebDev | Not measured | 1482 | Only MiniMax M3 measured |
| AA Coding Index | Not measured | 58.6 | Only MiniMax M3 measured |
| LiveCodeBench (Vals AI) | 63.1% | 82.2% | MiniMax M3 |
| SWE-bench Verified (Vals AI) | 60.2% | 75.0% | MiniMax M3 |
| Terminal-Bench 4 (Vals AI) | Not measured | 1.0% | Only MiniMax M3 measured |
| Vibe Code Bench (Vals AI) | Not measured | 47.6% | Only MiniMax M3 measured |
| Code migration (Vals AI) | Not measured | 19.9% | Only MiniMax M3 measured |
| Terminal-Bench 2.1 (Vals AI) | Not measured | 53.6% | Only MiniMax M3 measured |
| SWE-bench Multimodal (Vals AI) | Not measured | 33.3% | Only MiniMax M3 measured |
| Vibe Code Bench 1-100 (Vals AI) | Not measured | 9.2% | Only MiniMax M3 measured |
| SkillsBench (Vals AI) | Not measured | 51.5% | Only MiniMax M3 measured |
| CyberBench (Vals AI) | Not measured | 65.0% | Only MiniMax M3 measured |
| LiveCodeBench (Artificial Analysis run) | 85.3% | Not measured | Only Kimi K2 Thinking measured |
| SciCode | Not measured | 47.1% | Only MiniMax M3 measured |
| Terminal-Bench Hard (Artificial Analysis run) | 31.1% | 42.4% | MiniMax M3 |
| Terminal-Bench | 35.7% | Not measured | Only Kimi K2 Thinking measured |
| FrontierCode | Not measured | 14.7% | Only MiniMax M3 measured |
| WeirdML | 42.8% | Not measured | Only Kimi K2 Thinking measured |
| ALE-Bench | 597.5 | 640.0 | MiniMax M3 |
| LMArena Text: coding | Not measured | 1470 | Only MiniMax M3 measured |
| LMArena Text: software and IT | Not measured | 1466 | Only MiniMax M3 measured |
| LMArena WebDev: React | Not measured | 1478 | Only MiniMax M3 measured |
| LMArena WebDev: HTML | Not measured | 1455 | Only MiniMax M3 measured |
| LMArena WebDev: from an image | Not measured | 1470 | Only MiniMax M3 measured |
| Neither model has a result on the other 12 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Kimi K2 Thinking | MiniMax M3 | Leader |
|---|---|---|---|
| AA Intelligence Index | 22.0 | 29.2 | MiniMax M3 |
| Epoch Capabilities Index | 145.8 | 146.9 | MiniMax M3 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Kimi K2 Thinking | MiniMax M3 | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 29.5 | Only MiniMax M3 measured |
| Design Arena: agents building web apps | Not measured | 1200 | Only MiniMax M3 measured |
| Design Arena: agents building full-stack apps | Not measured | 1197 | Only MiniMax M3 measured |
| Design Arena: agents building mobile apps | Not measured | 1169 | Only MiniMax M3 measured |
| Design Arena: agents building Android apps | Not measured | 1164 | Only MiniMax M3 measured |
| Design Arena: agents building games | Not measured | 1158 | Only MiniMax M3 measured |
| Design Arena: HTML slides | Not measured | 1180 | Only MiniMax M3 measured |
| Design Arena: PowerPoint slides | Not measured | 1205 | Only MiniMax M3 measured |
| Design Arena: spreadsheets | Not measured | 1021 | Only MiniMax M3 measured |
| LiveBench research | Not measured | 68.0% | Only MiniMax M3 measured |
| LiveBench browser | Not measured | 8.0% | Only MiniMax M3 measured |
| LiveBench long agentic | Not measured | 12.4% | Only MiniMax M3 measured |
| Finance Agent (Vals AI) | Not measured | 48.3% | Only MiniMax M3 measured |
| Tax Agent Bench (Vals AI) | Not measured | 20.6% | Only MiniMax M3 measured |
| Legal Research Bench (Vals AI) | Not measured | 29.8% | Only MiniMax M3 measured |
| Excel Modeling Benchmark (Vals AI) | Not measured | 47.8% | Only MiniMax M3 measured |
| τ²-Bench Telecom | 93.0% | 88.9% | Kimi K2 Thinking |
| APEX-Agents | Not measured | 37.7% | Only MiniMax M3 measured |
| METR time horizons | 59.2% | Not measured | Only Kimi K2 Thinking measured |
| OSWorld 2.0 | Not measured | 4.6% | Only MiniMax M3 measured |
| Vending-Bench 2 | Not measured | $2,158 | Only MiniMax M3 measured |
| Neither model has a result on the other 18 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index03
| Benchmark | Kimi K2 Thinking | MiniMax M3 | Leader |
|---|---|---|---|
| GPQA Diamond | 78.5% | 92.7% | MiniMax M3 |
| LiveBench reasoning | Not measured | 74.5% | Only MiniMax M3 measured |
| MMMU Pro (Vals AI) | Not measured | 81.2% | Only MiniMax M3 measured |
| GPQA Diamond (Artificial Analysis run) | 83.8% | 92.9% | MiniMax M3 |
| Humanity's Last Exam (Artificial Analysis run) | 23.8% | 39.0% | MiniMax M3 |
| GPQA Diamond (Epoch AI run) | Not measured | 90.9% | Only MiniMax M3 measured |
| SimpleBench | Not measured | 45.8% | Only MiniMax M3 measured |
| Chess puzzles | Not measured | 13.0% | Only MiniMax M3 measured |
| Mystery game puzzles | Not measured | 8.0% | Only MiniMax M3 measured |
| LMArena Text: hard prompts | Not measured | 1447 | Only MiniMax M3 measured |
| LMArena Text: expert | Not measured | 1461 | Only MiniMax M3 measured |
| LMArena Text: hard prompts in English | Not measured | 1455 | Only MiniMax M3 measured |
| LMArena Vision: diagrams | Not measured | 1266 | Only MiniMax M3 measured |
| LMArena Vision: homework | Not measured | 1268 | Only MiniMax M3 measured |
| Neither model has a result on the other 7 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | Kimi K2 Thinking | MiniMax M3 | Leader |
|---|---|---|---|
| AIME | 85.4% | Not measured | Only Kimi K2 Thinking measured |
| MGSM (Vals AI) | 90.2% | Not measured | Only Kimi K2 Thinking measured |
| AA Math Index | 94.7 | Not measured | Only Kimi K2 Thinking measured |
| AIME 2025 (Artificial Analysis run) | 94.7% | Not measured | Only Kimi K2 Thinking measured |
| OTIS Mock AIME | Not measured | 71.1% | Only MiniMax M3 measured |
| ProofBench | Not measured | 18.0% | Only MiniMax M3 measured |
| LMArena Text: math | Not measured | 1429 | Only MiniMax M3 measured |
| LMArena Text: mathematical work | Not measured | 1448 | Only MiniMax M3 measured |
| Neither model has a result on the other 5 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index04
| Benchmark | Kimi K2 Thinking | MiniMax M3 | Leader |
|---|---|---|---|
| MMLU Pro | 81.1% | 84.2% | MiniMax M3 |
| LegalBench (Vals AI) | 80.2% | 85.4% | MiniMax M3 |
| CaseLaw (Vals AI) | 65.7% | Not measured | Only Kimi K2 Thinking measured |
| CorpFin (Vals AI) | 60.6% | 68.1% | MiniMax M3 |
| TaxEval (Vals AI) | 71.7% | 72.7% | MiniMax M3 |
| MortgageTax (Vals AI) | Not measured | 68.4% | Only MiniMax M3 measured |
| MedQA (Vals AI) | 92.6% | Not measured | Only Kimi K2 Thinking measured |
| MedCode (Vals AI) | Not measured | 46.3% | Only MiniMax M3 measured |
| MMLU Pro (Artificial Analysis run) | 84.8% | Not measured | Only Kimi K2 Thinking measured |
| LMArena Text: science | Not measured | 1452 | Only MiniMax M3 measured |
| LMArena Text: medicine and healthcare | Not measured | 1441 | Only MiniMax M3 measured |
| LMArena Text: legal and government | Not measured | 1437 | Only MiniMax M3 measured |
| LMArena Text: business and finance | Not measured | 1434 | Only MiniMax M3 measured |
| Neither model has a result on the other 4 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Kimi K2 Thinking | MiniMax M3 | Leader |
|---|---|---|---|
| Design Arena: websites | 1113 | 1257 | MiniMax M3 |
| Design Arena: UI components | Not measured | 1246 | Only MiniMax M3 measured |
| Design Arena: game dev | Not measured | 1210 | Only MiniMax M3 measured |
| Design Arena: data viz | Not measured | 1235 | Only MiniMax M3 measured |
| Design Arena: 3D | Not measured | 1209 | Only MiniMax M3 measured |
| Design Arena: all code categories | Not measured | 1247 | Only MiniMax M3 measured |
| Design Arena: SVG | Not measured | 1161 | Only MiniMax M3 measured |
| Design Arena: ASCII art | Not measured | 1161 | Only MiniMax M3 measured |
| LiveBench design | Not measured | 66.2% | Only MiniMax M3 measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Kimi K2 Thinking | MiniMax M3 | Leader |
|---|---|---|---|
| LiveBench data analysis | Not measured | 76.2% | Only MiniMax M3 measured |
| Public Benefits Bench (Vals AI) | Not measured | 64.1% | Only MiniMax M3 measured |
| SAGE (Vals AI) | Not measured | 50.6% | Only MiniMax M3 measured |
| MedScribe (Vals AI) | Not measured | 87.3% | Only MiniMax M3 measured |
| Vals Multimodal Index | Not measured | 60.0% | Only MiniMax M3 measured |
| IFBench | 68.1% | 82.9% | MiniMax M3 |
| Long context reasoning (AA-LCR) | 72.0% | 83.0% | MiniMax M3 |
| DTBench | Not measured | 78.9% | Only MiniMax M3 measured |
| LMCA | Not measured | 33.7% | Only MiniMax M3 measured |
| CL-bench | 17.6% | Not measured | Only Kimi K2 Thinking measured |
| Surface Evolver Bench | Not measured | 55.0% | Only MiniMax M3 measured |
| LMArena Text | Not measured | 1433 | Only MiniMax M3 measured |
| LMArena Text: instruction following | Not measured | 1433 | Only MiniMax M3 measured |
| LMArena Text: creative writing | Not measured | 1404 | Only MiniMax M3 measured |
| LMArena Vision | Not measured | 1251 | Only MiniMax M3 measured |
| LMArena Text: English | Not measured | 1443 | Only MiniMax M3 measured |
| LMArena Text: non-English | Not measured | 1420 | Only MiniMax M3 measured |
| LMArena Text: Chinese | Not measured | 1466 | Only MiniMax M3 measured |
| LMArena Text: German | Not measured | 1428 | Only MiniMax M3 measured |
| LMArena Text: Japanese | Not measured | 1380 | Only MiniMax M3 measured |
| LMArena Text: Korean | Not measured | 1373 | Only MiniMax M3 measured |
| LMArena Text: French | Not measured | 1444 | Only MiniMax M3 measured |
| LMArena Text: Spanish | Not measured | 1433 | Only MiniMax M3 measured |
| LMArena Text: Russian | Not measured | 1426 | Only MiniMax M3 measured |
| LMArena Text: Polish | Not measured | 1423 | Only MiniMax M3 measured |
| LMArena Text: multi-turn | Not measured | 1443 | Only MiniMax M3 measured |
| LMArena Text: longer queries | Not measured | 1446 | Only MiniMax M3 measured |
| LMArena Text: writing and language | Not measured | 1419 | Only MiniMax M3 measured |
| LMArena Text: entertainment and media | Not measured | 1403 | Only MiniMax M3 measured |
| LMArena Vision: English | Not measured | 1263 | Only MiniMax M3 measured |
| LMArena Vision: Chinese | Not measured | 1311 | Only MiniMax M3 measured |
| LMArena Vision: reading text | Not measured | 1259 | Only MiniMax M3 measured |
| LMArena Vision: humor | Not measured | 1235 | Only MiniMax M3 measured |
| LMArena Vision: recognizing things | Not measured | 1197 | Only MiniMax M3 measured |
| LMArena Vision: creative writing | Not measured | 1234 | Only MiniMax M3 measured |
| LMArena Document | Not measured | 1435 | Only MiniMax M3 measured |
| Neither model has a result on the other 10 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Kimi K2 Thinking | MiniMax M3 | 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
MiniMax M3 costs less
Repository review
50K input + 3K output tokens
MiniMax M3 costs less
Agent loop
200K input + 10K output tokens
MiniMax M3 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 | MoonshotAI Kimi K2 Thinking | MiniMax M3 |
|---|---|---|
| Maker | Moonshot AI | MiniMax |
| Weights | Open weights | Open weights |
| Released | November 6, 2025 | May 31, 2026 |
| Context window | 262K | 1.05M |
| Input $/Mtok | $0.60 | $0.30 |
| Output $/Mtok | $2.50 | $1.20 |
| Coding index | 63.4 | 57.8 |
| Independent measurements behind it | 1 | 2 |
| Cost per solved task | $0.61 | $0.56 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Kimi K2 Thinking better than MiniMax M3 for coding?
Pick Kimi K2 Thinking for coding: it scores 63.4 on the Vibe Coding Index against 57.8 for MiniMax M3, though on 1 benchmark against 2. Pick MiniMax M3 when output tokens cost 2.1x less, or when a solved task costs $0.56 against $0.61.
Which is cheaper, Kimi K2 Thinking or MiniMax M3?
Kimi K2 Thinking costs $0.60 per million input tokens and $2.50 per million output tokens; MiniMax M3 costs $0.30 and $1.20. Per solved SWE-bench Verified task, Kimi K2 Thinking costs $0.61 and MiniMax M3 $0.56.
Which has the bigger context window, Kimi K2 Thinking or MiniMax M3?
MiniMax M3, with 1.05M tokens against 262K.
How many benchmarks is the Kimi K2 Thinking vs MiniMax M3 comparison based on?
17 benchmarks have a result for both models as of October 21, 2026. Kimi K2 Thinking's coding index rests on 1 independent benchmark and MiniMax M3's on 2. Where only one model has a result, the page says so instead of naming a winner.