Compare / Gemini 3.1 Flash Lite Preview vs Kimi K2.5
Gemini 3.1 Flash Lite Preview vs Kimi K2.5
19 results from 3 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | Gemini 3.1 Flash Lite Preview | Kimi K2.5 |
|---|---|---|---|
| Arena (formerly LMArena)reported | OverallArena rating | 1,433 | 1,445 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,421 | 1,441 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,413 | 1,423 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,443 | 1,483 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,407 | 1,439 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,433 | 1,450 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,420 | 1,428 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 34.7 | 46.8 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 15.6 | 23.5 |
| Artificial Analysisreported | GPQA Diamond% of questions | 82.2% | 87.9% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 17.2% | 30.7% |
| Artificial Analysisreported | IFBench% of instructions | 77.2% | 70.2% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 74.3% | 78.0% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 31.1% | 45.7% |
| Artificial Analysisreported | Terminal-Bench Hard% of tasks | 24.2% | 34.8% |
| Artificial Analysisreported | Τ-bench banking% of tasks | 9.7% | 14.2% |
| Artificial Analysisreported | Τ²-bench telecom% of tasks | 31.3% | 95.9% |
| Vectara Hallucination Leaderboardreported | Answer rate% of documents | 99.6% | 92.2% |
| Vectara Hallucination Leaderboardreported | Hallucination rate% of summaries | 8.2% | 14.2% |
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.