Kimi K3 vs Hy4 preview
Kimi K3 leads Hy4 preview on coding, 62.3 to 44.3 on the Vibe Coding Index, and Hy4 preview costs 6.7x less per output token ($2.2509 vs $15.00 per million). Updated October 21, 2026; 23 benchmarks have a result for both.
Kimi K3 or Hy4 preview: which to pick
Pick Kimi K3 for coding: it scores 62.3 on the Vibe Coding Index against 44.3 for Hy4 preview.
Pick Hy4 preview for output tokens 6.7x cheaper ($2.2509 vs $15.00 per million) and work measured by Code migration (Vals AI) or IOI (Vals AI), where it leads.
- Shared results
- 23
- Kimi K3 only
- 94
- Hy4 preview only
- 1
- Like-for-like categories
- 5 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.
Coding
Vibe Coding Index, six public coding benchmarks
Kimi K3
Kimi K3 scores 62.3 against 44.3.
6 and 3 benchmarks behind the scores
Intelligence
AA Intelligence Index
No call
Only Kimi K3 has a AA Intelligence Index result, 43.6.
Directional only
Agentic work
AA Agentic Index
No call
Only Kimi K3 has a AA Agentic Index result, 50.0.
Directional only
Reasoning
GPQA Diamond
No call
Only Kimi K3 has a GPQA Diamond result, 92.9%.
Directional only
Knowledge
MMLU Pro
No call
Only Kimi K3 has a MMLU Pro result, 88.0%.
Directional only
Web design
Design Arena: websites
No call
Only Kimi K3 has a Design Arena: websites result, 1335.
Directional only
General
Vals Index
Hy4 preview
Hy4 preview leads, 53.2% against 53.0%.
Same board, both measured
Long documents
Listed context window
No call
Both take 1.05M tokens.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Hy4 preview
Kimi K3 $0.011, Hy4 preview $0.0019 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Kimi K3 has the number, $0.81.
Directional only
At a glance
- Coding index
- Kimi K3 62.3 from 6 benchmarks, Hy4 preview 44.3 from 3 benchmarks.
- Widest category
- Math, 12.0 points apart on average over 1 shared benchmark.
- Widest single gap
- Code migration (Vals AI): Kimi K3 16.1% against Hy4 preview 47.4%, 31.3 points apart on a 0 to 100 scale.
- Output price
- Kimi K3 $15.00 against Hy4 preview $2.2509 per million output tokens, 6.7x 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.
| Category | Kimi K3 | Bars | Hy4 preview | Ties |
|---|---|---|---|---|
| Coding | 8 | 2 | ||
| Agentic work | 5 | 1 | 1 tie | |
| Math | 1 | 0 | ||
| Knowledge | 2 | 0 | ||
| General | 1 | 2 | ||
| All shared boards | 17 | 5 | 1 tie |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index82
| Benchmark | Kimi K3 | Hy4 preview | Leader |
|---|---|---|---|
| LiveBench coding | 59.3% | Not measured | Only Kimi K3 measured |
| LMArena WebDev | 1655 | 1631 | Kimi K3 |
| AA Coding Index | 76.2 | Not measured | Only Kimi K3 measured |
| LiveCodeBench (Vals AI) | 87.2% | Not measured | Only Kimi K3 measured |
| SWE-bench Verified (Vals AI) | 93.4% | Not measured | Only Kimi K3 measured |
| Terminal-Bench 4 (Vals AI) | 17.2% | 8.1% | Kimi K3 |
| Vibe Code Bench (Vals AI) | 85.0% | 77.5% | Kimi K3 |
| Code migration (Vals AI) | 16.1% | 47.4% | Hy4 preview |
| ProgramBench (Vals AI) | 2.0% | 0.0% | Kimi K3 |
| IOI (Vals AI) | 48.9% | 59.3% | Hy4 preview |
| Terminal-Bench 2.1 (Vals AI) | 80.9% | 55.1% | Kimi K3 |
| SWE-bench Multimodal (Vals AI) | 44.2% | Not measured | Only Kimi K3 measured |
| Vibe Code Bench 1-100 (Vals AI) | 18.2% | Not measured | Only Kimi K3 measured |
| CyberBench (Vals AI) | 75.2% | 65.4% | Kimi K3 |
| SciCode | 59.5% | Not measured | Only Kimi K3 measured |
| FrontierCode | 44.2% | Not measured | Only Kimi K3 measured |
| FrontierSWE | 25.9% | Not measured | Only Kimi K3 measured |
| WeirdML | 82.6% | Not measured | Only Kimi K3 measured |
| ALE-Bench | 1524.5 | Not measured | Only Kimi K3 measured |
| LMArena Text: coding | 1508 | Not measured | Only Kimi K3 measured |
| LMArena Text: software and IT | 1499 | Not measured | Only Kimi K3 measured |
| LMArena WebDev: React | 1670 | 1653 | Kimi K3 |
| LMArena WebDev: HTML | 1644 | 1620 | Kimi K3 |
| LMArena WebDev: from an image | 1579 | Not measured | Only Kimi K3 measured |
| Neither model has a result on the other 15 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | Kimi K3 | Hy4 preview | Leader |
|---|---|---|---|
| AA Intelligence Index | 43.6 | Not measured | Only Kimi K3 measured |
| Epoch Capabilities Index | 157.9 | Not measured | Only Kimi K3 measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index51
| Benchmark | Kimi K3 | Hy4 preview | Leader |
|---|---|---|---|
| AA Agentic Index | 50.0 | Not measured | Only Kimi K3 measured |
| Design Arena: agents building web apps | 1295 | Not measured | Only Kimi K3 measured |
| Design Arena: agents building full-stack apps | 1315 | Not measured | Only Kimi K3 measured |
| Design Arena: agents building mobile apps | 1269 | Not measured | Only Kimi K3 measured |
| Design Arena: agents building Android apps | 1250 | Not measured | Only Kimi K3 measured |
| Design Arena: agents building Godot games | 1223 | Not measured | Only Kimi K3 measured |
| Design Arena: agents building games | 1252 | Not measured | Only Kimi K3 measured |
| Design Arena: HTML slides | 1256 | Not measured | Only Kimi K3 measured |
| Design Arena: PowerPoint slides | 1273 | Not measured | Only Kimi K3 measured |
| Design Arena: spreadsheets | 1058 | Not measured | Only Kimi K3 measured |
| LiveBench research | 79.0% | Not measured | Only Kimi K3 measured |
| LiveBench browser | 16.0% | Not measured | Only Kimi K3 measured |
| LiveBench long agentic | 42.8% | Not measured | Only Kimi K3 measured |
| SREBench (Vals AI) | Not measured | 2.3% | Only Hy4 preview measured |
| Finance Agent (Vals AI) | 53.1% | 55.1% | Hy4 preview |
| Tax Agent Bench (Vals AI) | 33.4% | 32.6% | Kimi K3 |
| Legal Research Bench (Vals AI) | 46.1% | 45.2% | Kimi K3 |
| Harvey's Legal Agent Benchmark (Vals AI) | 12.9% | 8.8% | Kimi K3 |
| Excel Modeling Benchmark (Vals AI) | 66.7% | 58.1% | Kimi K3 |
| BioMysteryBench (Vals AI) | 72.6% | 69.3% | Kimi K3 |
| Terminal-Bench-Science (Vals AI) | 1.4% | 1.4% | Tie |
| Time Horizon Index (Vals AI) | 10.5% | Not measured | Only Kimi K3 measured |
| APEX-Agents | 47.8% | Not measured | Only Kimi K3 measured |
| PostTrainBench | 32.0% | Not measured | Only Kimi K3 measured |
| Vending-Bench 2 | $5,165 | Not measured | Only Kimi K3 measured |
| Neither model has a result on the other 16 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index14 boards, none shared
| Benchmark | Kimi K3 | Hy4 preview | Leader |
|---|---|---|---|
| GPQA Diamond | 92.9% | Not measured | Only Kimi K3 measured |
| LiveBench reasoning | 90.7% | Not measured | Only Kimi K3 measured |
| MMMU Pro (Vals AI) | 88.2% | Not measured | Only Kimi K3 measured |
| GPQA Diamond (Artificial Analysis run) | 93.5% | Not measured | Only Kimi K3 measured |
| Humanity's Last Exam (Artificial Analysis run) | 46.9% | Not measured | Only Kimi K3 measured |
| GPQA Diamond (Epoch AI run) | 93.1% | Not measured | Only Kimi K3 measured |
| ARC-AGI | 94.5% | Not measured | Only Kimi K3 measured |
| ARC-AGI-2 | 60.4% | Not measured | Only Kimi K3 measured |
| SimpleBench | 60.7% | Not measured | Only Kimi K3 measured |
| Chess puzzles | 37.0% | Not measured | Only Kimi K3 measured |
| Mystery game puzzles | 26.0% | Not measured | Only Kimi K3 measured |
| LMArena Text: hard prompts | 1495 | Not measured | Only Kimi K3 measured |
| LMArena Text: expert | 1520 | Not measured | Only Kimi K3 measured |
| LMArena Text: hard prompts in English | 1496 | Not measured | Only Kimi K3 measured |
| Neither model has a result on the other 8 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Kimi K3 | Hy4 preview | Leader |
|---|---|---|---|
| OTIS Mock AIME | 97.2% | Not measured | Only Kimi K3 measured |
| FrontierMath, tiers 1 to 3 | 72.2% | Not measured | Only Kimi K3 measured |
| FrontierMath, tier 4 | 39.0% | Not measured | Only Kimi K3 measured |
| ProofBench | 87.0% | 75.0% | Kimi K3 |
| LMArena Text: math | 1494 | Not measured | Only Kimi K3 measured |
| LMArena Text: mathematical work | 1505 | Not measured | Only Kimi K3 measured |
| Neither model has a result on the other 7 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | Kimi K3 | Hy4 preview | Leader |
|---|---|---|---|
| MMLU Pro | 88.0% | Not measured | Only Kimi K3 measured |
| LegalBench (Vals AI) | 86.2% | 83.8% | Kimi K3 |
| CorpFin (Vals AI) | 71.6% | Not measured | Only Kimi K3 measured |
| TaxEval (Vals AI) | 75.7% | Not measured | Only Kimi K3 measured |
| MortgageTax (Vals AI) | 66.3% | Not measured | Only Kimi K3 measured |
| MedCode (Vals AI) | 49.4% | 43.3% | Kimi K3 |
| SimpleQA Verified | 50.6% | Not measured | Only Kimi K3 measured |
| LMArena Text: science | 1508 | Not measured | Only Kimi K3 measured |
| LMArena Text: medicine and healthcare | 1489 | Not measured | Only Kimi K3 measured |
| LMArena Text: legal and government | 1503 | Not measured | Only Kimi K3 measured |
| LMArena Text: business and finance | 1470 | Not measured | Only Kimi K3 measured |
| 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 | Kimi K3 | Hy4 preview | Leader |
|---|---|---|---|
| Design Arena: websites | 1335 | Not measured | Only Kimi K3 measured |
| Design Arena: UI components | 1344 | Not measured | Only Kimi K3 measured |
| Design Arena: game dev | 1359 | Not measured | Only Kimi K3 measured |
| Design Arena: data viz | 1351 | Not measured | Only Kimi K3 measured |
| Design Arena: 3D | 1388 | Not measured | Only Kimi K3 measured |
| Design Arena: all code categories | 1360 | Not measured | Only Kimi K3 measured |
| Design Arena: SVG | 1302 | Not measured | Only Kimi K3 measured |
| LiveBench design | 77.2% | Not measured | Only Kimi K3 measured |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index12
| Benchmark | Kimi K3 | Hy4 preview | Leader |
|---|---|---|---|
| Vals Index | 53.0% | 53.2% | Hy4 preview |
| LiveBench data analysis | 78.7% | Not measured | Only Kimi K3 measured |
| Public Benefits Bench (Vals AI) | 68.2% | 68.6% | Hy4 preview |
| SAGE (Vals AI) | 52.8% | Not measured | Only Kimi K3 measured |
| MedScribe (Vals AI) | 88.0% | 83.6% | Kimi K3 |
| Vals Multimodal Index | 73.4% | Not measured | Only Kimi K3 measured |
| Long context reasoning (AA-LCR) | 88.7% | Not measured | Only Kimi K3 measured |
| DTBench | 91.2% | Not measured | Only Kimi K3 measured |
| LMCA | 52.7% | Not measured | Only Kimi K3 measured |
| Surface Evolver Bench | 95.0% | Not measured | Only Kimi K3 measured |
| Furniture assembly | 34.2% | Not measured | Only Kimi K3 measured |
| LMArena Text | 1476 | Not measured | Only Kimi K3 measured |
| LMArena Text: instruction following | 1485 | Not measured | Only Kimi K3 measured |
| LMArena Text: creative writing | 1457 | Not measured | Only Kimi K3 measured |
| LMArena Text: English | 1481 | Not measured | Only Kimi K3 measured |
| LMArena Text: non-English | 1466 | Not measured | Only Kimi K3 measured |
| LMArena Text: Chinese | 1529 | Not measured | Only Kimi K3 measured |
| LMArena Text: German | 1490 | Not measured | Only Kimi K3 measured |
| LMArena Text: Japanese | 1494 | Not measured | Only Kimi K3 measured |
| LMArena Text: Korean | 1460 | Not measured | Only Kimi K3 measured |
| LMArena Text: French | 1492 | Not measured | Only Kimi K3 measured |
| LMArena Text: Spanish | 1472 | Not measured | Only Kimi K3 measured |
| LMArena Text: Russian | 1485 | Not measured | Only Kimi K3 measured |
| LMArena Text: Polish | 1458 | Not measured | Only Kimi K3 measured |
| LMArena Text: multi-turn | 1489 | Not measured | Only Kimi K3 measured |
| LMArena Text: longer queries | 1494 | Not measured | Only Kimi K3 measured |
| LMArena Text: writing and language | 1471 | Not measured | Only Kimi K3 measured |
| LMArena Text: entertainment and media | 1453 | Not measured | Only Kimi K3 measured |
| Neither model has a result on the other 18 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Kimi K3 | Hy4 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
Hy4 preview costs less
Repository review
50K input + 3K output tokens
Hy4 preview costs less
Agent loop
200K input + 10K output tokens
Hy4 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 | MoonshotAI Kimi K3 | Tencent Hy4 preview |
|---|---|---|
| Maker | Moonshot AI | Tencent |
| Weights | Open weights | Open weights |
| Released | July 16, 2026 | August 28, 2026 |
| Context window | 1.05M | 1.05M |
| Input $/Mtok | $3.00 | $0.7506 |
| Output $/Mtok | $15.00 | $2.2509 |
| Coding index | 62.3 | 44.3 |
| Independent measurements behind it | 6 | 3 |
| Cost per solved task | $0.81 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Kimi K3 better than Hy4 preview for coding?
Pick Kimi K3 for coding: it scores 62.3 on the Vibe Coding Index against 44.3 for Hy4 preview. Pick Hy4 preview for output tokens 6.7x cheaper ($2.2509 vs $15.00 per million) and work measured by Code migration (Vals AI) or IOI (Vals AI), where it leads.
Which is cheaper, Kimi K3 or Hy4 preview?
Kimi K3 costs $3.00 per million input tokens and $15.00 per million output tokens; Hy4 preview costs $0.7506 and $2.2509. Cost per solved task is known for Kimi K3 only.
Which has the bigger context window, Kimi K3 or Hy4 preview?
Neither: both take 1.05M tokens.
How many benchmarks is the Kimi K3 vs Hy4 preview comparison based on?
23 benchmarks have a result for both models as of October 21, 2026. Kimi K3's coding index rests on 6 independent benchmarks and Hy4 preview's on 3. Where only one model has a result, the page says so instead of naming a winner.