Claude Haiku 5.5 vs 
Kimi K3
Both are SOTA right now. Each row is one board’s call, and the marked figure is the one that board places higher. No row is added to another, so there is no overall winner: the boards measure different things, and where they split, that is the finding.Each rank is the board’s own: the rank it publishes, or the model’s place in the board’s own order where it publishes none. Scores from different boards cannot be compared.
| Board | Claude Haiku 5.5 | Kimi K3 |
|---|---|---|
| Artificial Analysis LLM Leaderboard Artificial Analysis Intelligence Index | 43.4 No. 19 of 375 | 43.6 No. 18 of 375 |
| Arena — Agent Overall | — | 0.038 No. 17 of 52 |
| Artificial Analysis Coding Agents Artificial Analysis Coding Agent Index | 0.414 No. 22 of 24 | 0.519 No. 15 of 24 |
| Epoch AI Benchmarking Hub FrontierMath Tiers 1–3 | 75.1% No. 18 of 82 | 72.2% No. 22 of 82 |
| HarnessTax — SWE-bench Lite SWE-bench Lite | — | 76.7% No. 4 of 7 |
| HarnessTax — Terminal-Bench 2.0 Terminal-Bench 2.0 | — | 73.3% No. 4 of 7 |
| Arena — Code Overall | 1587 No. 30 of 142 | 1654 No. 14 of 142 |
| Arena — Text Overall (style control) | — | 1488 No. 16 of 414 |
At a glance
| Developer | Anthropic | Kimi |
|---|---|---|
| Weights | closed | open |
| Released | — | 2026-07-16 |
| Price Blended Price, USD per 1M tokens (3:1 input:output) | $0.20 | $6.00 |
| Value: where it sits on the kill line | Executioner | SOTA |
Claude Haiku 5.5 →Kimi K3 → How the kill line is drawn →