Up to four models
GPT-5.4 vs Kimi K2.5
On the index (GPT-4 = 100), GPT-5.4 scores 166 and Kimi K2.5 145. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
GPT-5.4
166150–182
WeightsClosed
ReleasedMar 2026
Price in / out$2.5 / $15
Context1.05M
Core tests taken7 of 9
Kimi K2.5
145130–161
WeightsOpen
ReleasedJan 2026
Price in / out$0.50 / $2.5
Context262K
Core tests taken5 of 9
GPT-5.4 leads Kimi K2.5 by 21 points, ahead on 4 of 4 shared core tests, at 5.6× the price.
Test by testthird-party results
GPT-5.4Kimi K2.5
Every result as a table
| Test | GPT-5.4 | Kimi K2.5 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 93.3% | 87.6% |
| Humanity's Last Exam | 36.2% | 24.4% |
| GPQA Diamond (Vals AI) | 91.7% | 84.1% |
| MMLU Pro | 87.5% | 85.9% |
| MMMU Pro | 87.5% | 84.3% |
| CL-bench | 27.9% | 19.3% |
| SimpleQA Verified | 45.1% | 34.3% |
| Maths | ||
| AIME-style maths (OTIS mock) | 97.8% | 92.2% |
| FrontierMath (2025 set) | 47.6% | 27.9% |
| FrontierMath tier 4 (2025 set) | 27.1% | 4.2% |
| Coding | ||
| SWE-bench Verified | 76.9% | 73.8% |
| Code Migration | 35.0% | 7.0% |
| LiveCodeBench | 84.1% | 83.9% |
| SWE-bench | 78.2% | 70.0% |
| Vibe Code Bench v1.1 | 67.4% | 17.5% |
| WeirdML | 77.7% | 45.6% |
| Professional work | ||
| CorpFin v2 | 65.3% | 68.3% |
| MedCode | 41.3% | 39.3% |
| MedScribe | 77.5% | 76.4% |
| MortgageTax | 68.3% | 66.5% |
| TaxEval v2 | 74.0% | 74.2% |
| Novel problems | ||
| ARC-AGI-2 | 74.0% | 11.8% |
| ARC-AGI-1 | 93.7% | 65.3% |
Task length · METR45 sec
Tests at least two of these models have taken. Bold rows are the core tests.