Compare / Qwen3 Coder 480B A35B vs Kimi K2.5
Qwen3 Coder 480B A35B vs Kimi K2.5
20 results from 3 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | Qwen3 Coder 480B A35B | Kimi K2.5 |
|---|---|---|---|
| Arena (formerly LMArena)reported | OverallArena rating | 1,401 | 1,445 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,391 | 1,441 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,362 | 1,423 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,379 | 1,483 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,384 | 1,439 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,387 | 1,450 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,366 | 1,428 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 11.9 | 23.5 |
| Artificial Analysisreported | GPQA Diamond% of questions | 61.8% | 87.9% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 4.5% | 30.7% |
| Artificial Analysisreported | IFBench% of instructions | 40.5% | 70.2% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 45.7% | 78.0% |
| Artificial Analysisreported | Terminal-Bench Hard% of tasks | 18.9% | 34.8% |
| Artificial Analysisreported | Τ²-bench telecom% of tasks | 43.6% | 95.9% |
| OpenHands Indexreported | Average% resolved | 30.9% | 49.2% |
| OpenHands Indexreported | Front end (SWE-Bench Multimodal)% resolved | 23.5% | 32.8% |
| OpenHands Indexreported | Greenfield (Commit0)% resolved | 0.0% | 18.8% |
| OpenHands Indexreported | Information gathering (GAIA)% resolved | 33.9% | 63.6% |
| OpenHands Indexreported | Issue resolution (SWE-Bench)% resolved | 62.4% | 68.8% |
| OpenHands Indexreported | Testing (SWT-Bench)% resolved | 34.9% | 61.9% |
Bold green marks the better value on that metric. Where a source reports ranges that overlap, the difference may not be meaningful; see the benchmark page for ranges.