Up to four models
DeepSeek-V4-Pro vs GPT-5.6 Luna
On the index (GPT-4 = 100), DeepSeek-V4-Pro scores 163 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.
163147–178
WeightsOpen
ReleasedAug 2026
Price in / out$0.11 / $4.44
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
DeepSeek-V4-Pro and GPT-5.6 Luna are too close to call (0 points apart, within the uncertainty), ahead on 4 of 6 shared core tests, at 2.7× the price.
Test by testthird-party results
Every result as a table
| Test | DeepSeek-V4-Pro | GPT-5.6 Luna |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 91.7% | 91.6% |
| LiveBench Data Analysis | 79.2% | 78.0% |
| LiveBench IF | 67.7% | 60.1% |
| LiveBench Language | 82.1% | 72.6% |
| LiveBench Reasoning | 85.8% | 85.6% |
| GPQA Diamond (Vals AI) | 92.4% | 91.7% |
| MMLU Pro | 87.0% | 86.0% |
| SimpleQA Verified | 52.9% | 41.0% |
| Maths | ||
| FrontierMath (tiers 1-3) | 64.6% | 82.1% |
| AIME-style maths (OTIS mock) | 98.6% | 98.3% |
| LiveBench Mathematics | 95.1% | 87.2% |
| ProofBench v1.1 | 50.0% | 60.0% |
| Chess puzzles | 46.0% | 43.0% |
| FrontierMath (tier 4) | 26.8% | 61.0% |
| Coding | ||
| Terminal-Bench 4.0 | 14.1% | 11.6% |
| LiveBench Agentic Coding | 54.9% | 48.4% |
| LiveBench Coding | 77.2% | 82.9% |
| Code Migration | 41.5% | 44.5% |
| CyberBench v1.1 | 72.9% | 73.6% |
| IOI | 51.6% | 61.8% |
| ProgramBench | 0.0% | 0.0% |
| SWE-bench | 96.4% | 93.0% |
| Vibe Code Bench 1-100 | 17.5% | 22.6% |
| Vibe Code Bench v1.1 | 82.3% | 77.1% |
| MirrorCode | 33.3% | 21.1% |
| WeirdML | 66.2% | 60.9% |
| Agents | ||
| APEX-Agents | 47.3% | 43.0% |
| SkillsBench | 53.8% | 60.5% |
| SRE Bench | 1.9% | 1.5% |
| Terminal-Bench-Science | 4.3% | 0.0% |
| DTBench | 93.9% | 89.1% |
| LMCA | 45.5% | 48.5% |
| Mystery game puzzles | 43.0% | 21.0% |
| Professional work | ||
| Finance Agent (v2) | 50.4% | 55.0% |
| CorpFin v2 | 65.4% | 64.2% |
| EMB | 52.8% | 67.1% |
| LegalBench | 82.4% | 84.0% |
| Legal Research Bench | 40.9% | 36.5% |
| MedCode | 42.5% | 42.4% |
| MedScribe | 80.2% | 84.4% |
| Tax Agent Bench | 26.8% | 24.8% |
| TaxEval v2 | 73.1% | 76.2% |
| Novel problems | ||
| ARC-AGI-2 | 61.3% | 59.6% |
| ARC-AGI-1 | 90.5% | 90.7% |
Tests at least two of these models have taken. Bold rows are the core tests.