Up to four models
Kimi K2.5 vs Gemini 3.1 Flash-Lite
On the index (GPT-4 = 100), Kimi K2.5 scores 145 and Gemini 3.1 Flash-Lite 139. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Kimi K2.5
145130–161
WeightsOpen
ReleasedJan 2026
Price in / out$0.50 / $2.5
Context262K
Core tests taken5 of 9
139122–155
WeightsClosed
ReleasedMar 2026
Price in / out$0.25 / $1.5
Context1.05M
Core tests taken4 of 9
Kimi K2.5 leads Gemini 3.1 Flash-Lite by 7 points, ahead on 3 of 3 shared core tests, at 1.8× the price.
Test by testthird-party results
Kimi K2.5
Gemini 3.1 Flash-Lite
Every result as a table
| Test | Kimi K2.5 | Gemini 3.1 Flash-Lite |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 87.6% | 81.8% |
| Humanity's Last Exam | 24.4% | 8.6% |
| GPQA Diamond (Vals AI) | 84.1% | 81.1% |
| MMLU Pro | 85.9% | 86.2% |
| MMMU Pro | 84.3% | 82.5% |
| Maths | ||
| AIME-style maths (OTIS mock) | 92.2% | 80.0% |
| Coding | ||
| Code Migration | 7.0% | 4.6% |
| LiveCodeBench | 83.9% | 80.1% |
| SWE-bench | 70.0% | 62.8% |
| Vibe Code Bench v1.1 | 17.5% | 0.0% |
| WeirdML | 45.6% | 52.2% |
| Professional work | ||
| Finance Agent (v2) | 35.8% | 30.0% |
| CorpFin v2 | 68.3% | 59.4% |
| EMB | 28.5% | 8.6% |
| Legal Research Bench | 15.9% | 3.4% |
| MedCode | 39.3% | 47.6% |
| MedScribe | 76.4% | 63.9% |
| MortgageTax | 66.5% | 68.0% |
| TaxEval v2 | 74.2% | 71.8% |
Tests at least two of these models have taken. Bold rows are the core tests.