Up to four models
Kimi K3 vs GPT-5.6 Luna
On the index (GPT-4 = 100), Kimi K3 scores 168 and GPT-5.6 Luna 162. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Kimi K3
168152–183
WeightsOpen
ReleasedJul 2026
Price in / out$0.19 / $16
Context1.05M
Core tests taken6 of 9
GPT-5.6 Luna
162146–178
WeightsClosed
ReleasedJul 2026
Price in / out$0.20 / $1.2
Context1.05M
Core tests taken6 of 9
Kimi K3 leads GPT-5.6 Luna by 5 points, ahead on 4 of 6 shared core tests, at 9.2× the price.
Test by testthird-party results
Kimi K3GPT-5.6 Luna
Every result as a table
| Test | Kimi K3 | GPT-5.6 Luna |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 93.1% | 91.6% |
| LiveBench Data Analysis | 78.7% | 78.0% |
| LiveBench IF | 71.4% | 60.1% |
| LiveBench Language | 85.5% | 72.6% |
| LiveBench Reasoning | 90.7% | 85.6% |
| GPQA Diamond (Vals AI) | 92.9% | 91.7% |
| MMLU Pro | 88.0% | 86.0% |
| MMMU Pro | 88.1% | 85.0% |
| SimpleBench | 60.7% | 46.8% |
| SimpleQA Verified | 50.6% | 41.0% |
| Maths | ||
| FrontierMath (tiers 1-3) | 72.2% | 82.1% |
| AIME-style maths (OTIS mock) | 97.2% | 98.3% |
| LiveBench Mathematics | 84.4% | 87.2% |
| ProofBench v1.1 | 87.0% | 60.0% |
| Chess puzzles | 37.0% | 43.0% |
| FrontierMath (tier 4) | 39.0% | 61.0% |
| Surface Evolver Bench | 95.0% | 61.9% |
| Coding | ||
| Terminal-Bench 4.0 | 17.2% | 11.6% |
| LiveBench Agentic Coding | 62.2% | 48.4% |
| LiveBench Coding | 81.5% | 82.9% |
| Code Migration | 16.1% | 44.5% |
| CyberBench v1.1 | 75.2% | 73.6% |
| IOI | 48.9% | 61.8% |
| ProgramBench | 2.0% | 0.0% |
| SWE-bench | 93.4% | 93.0% |
| Vibe Code Bench 1-100 | 18.2% | 22.6% |
| Vibe Code Bench v1.1 | 85.0% | 77.1% |
| DeepSWE | 68.5% | 67.2% |
| MirrorCode | 54.4% | 21.1% |
| WeirdML | 82.6% | 60.9% |
| Agents | ||
| APEX-Agents | 47.8% | 43.0% |
| BioMysteryBench | 72.6% | 61.5% |
| Terminal-Bench-Science | 1.4% | 0.0% |
| DTBench | 91.2% | 89.1% |
| Furniture assembly | 34.2% | 42.5% |
| LMCA | 52.7% | 48.5% |
| Mystery game puzzles | 26.0% | 21.0% |
| Professional work | ||
| Finance Agent (v2) | 53.1% | 55.0% |
| CorpFin v2 | 71.6% | 64.2% |
| EMB | 66.7% | 67.1% |
| LegalBench | 86.2% | 84.0% |
| Legal Research Bench | 46.2% | 36.5% |
| MedCode | 49.4% | 42.4% |
| MedScribe | 88.0% | 84.4% |
| MortgageTax | 66.3% | 67.3% |
| Public Benefits Bench v1.1 | 68.2% | 61.2% |
| Tax Agent Bench | 33.4% | 24.8% |
| TaxEval v2 | 75.7% | 76.2% |
| Novel problems | ||
| ARC-AGI-2 | 60.4% | 59.6% |
| ARC-AGI-1 | 94.5% | 90.7% |
Tests at least two of these models have taken. Bold rows are the core tests.