Kimi K3 vs Step 5 Preview
Kimi K3 
Step 5 Preview Verdict
- Kimi K3 (max) leads on quality: 67.2 vs 65.0.
- Kimi K3 (max) is stronger in agents & tools, coding, composite, human preference, long context, maths, multimodal.
- Step 5 Preview is stronger in knowledge, reasoning.
- Step 5 Preview is 4.2× cheaper ($1.43 vs $6.00 per 1M blended).
| Metric | Kimi K3 (max) | Step 5 Preview |
|---|---|---|
| BenchLeader Index | 67.2 | 65.0 |
| Agents & tools score | 67.5 | – |
| Coding score | 67.0 | – |
| Composite score | 84.5 | 84.4 |
| Human preference score | 68.8 | – |
| Knowledge score | 64.5 | 72.0 |
| Long context score | 71.0 | 70.8 |
| Maths score | 65.4 | – |
| Multimodal score | 65.1 | 60.8 |
| Reasoning score | 68.6 | 82.1 |
| Blended price $/M | $6.00 | $1.43 |
| Output speed | 37 tok/s | – |
| Time to first answer | 58.3 s | – |
| Context window | 1.0M | 1M |
| GPQA Diamond | 93.1% | – |
| FrontierMath Tiers 1–3 | 72.2% | – |
| FrontierMath Tier 4 | 39.0% | – |
| OTIS Mock AIME | 97.2% | – |
| SimpleQA Verified | 50.6% | – |
| SimpleBench | 60.7% | – |
| SciCode | 58.7% | – |
| WeirdML | 82.6% | – |
| LMArena Text | 1485 | – |
| LMArena Hard Prompts | 1514 | – |
| LMArena Coding | 1538 | – |
| LMArena WebDev | 1674 | – |
| LMArena Agent | 6.2 | – |
| AA Intelligence Index | 43.8 | 43.6 |
| AA-LCR | 88.7% | 88.3% |
| MMMU-Pro | 80.5% | 76.4% |
| AA-Omniscience | 19.7 | 16.4 |
| GPQA Diamond (AA) | 93.5% | – |
| Humanity's Last Exam (AA) | 46.9% | 46.5% |
| SciCode (AA) | 59.5% | 58.9% |
| LiveCodeBench | 87.2% | – |
| MMLU-Pro | 88.0% | – |
| IOI | 48.9% | – |
| MCP Atlas | 82.3% | – |
| ARC-AGI-1 | 94.5% | – |
| ARC-AGI-2 | 60.4% | – |
| CritPt | 23.4% | 20.9% |
| GDPval (AA) | 52.4% | 53.6% |
| τ²-Bench Banking (AA) | 46.0% | – |
| ITBench SRE (AA) | 47.7% | – |
| Analyst Agent (AA) | 38.8% | – |
| APEX-Agents (AA) | 41.3% | – |
| BioMysteryBench | 71.5% | – |
| Code Migration | 16.1% | – |
| Excel Modeling Benchmark | 66.4% | – |
| Tax Agent Bench | 68.7% | – |
| Terminal-Bench 4.0 (Vals) | 12.6% | – |
| Terminal-Bench Science | 2.9% | – |
| Time Horizon Index: KSP | 10.5% | – |
| Vibe Code Bench 1-100 | 18.2% | – |
| FORTRESS | 26.6% | – |
| LMArena Maths | 1501 | – |
| LMArena Creative Writing | 1458 | – |
| LMArena Instruction Following | 1484 | – |
| LMArena Multi-turn | 1496 | – |
| LMArena Longer Queries | 1500 | – |
| Chess Puzzles | 39.0% | – |
| Mystery Game Puzzles | 26.0% | – |
| Surface Evolver Bench | 93.0% | – |
| DeepSWE | 68.5% | – |
| LMCA | 52.7% | – |
| DTBench | 91.2% | – |
| CursorBench | 60.8% | – |
| ALE-Bench | 1524.5 | – |
| GDP.pdf | 19.0% | – |
| FrontierSWE | 25.9% | – |
Data as of 2026-09-19. Best configuration of each model; every score links to its source on the model pages.
Kimi K3 vs Step 5 Preview: questions
- Is Kimi K3 better than Step 5 Preview?
- Kimi K3 (max) leads on quality: 67.2 vs 65.0. The BenchLeader Index combines every independent quality benchmark; Kimi K3 (max) is ahead overall as of 2026-09-19, but check the category scores for your use.
- Which is cheaper, Kimi K3 or Step 5 Preview?
- Step 5 Preview is cheaper: $1.43 against $6.00 per million tokens, blended at three input tokens per output token.
- Which has the larger context window?
- Kimi K3 accepts more context: 1.0M against 1M tokens.