Up to four models
Claude Opus 4.6 vs DeepSeek-V4.1-Flash
On the index (GPT-4 = 100), Claude Opus 4.6 scores 165 and DeepSeek-V4.1-Flash 163. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Claude Opus 4.6
165149–181
WeightsClosed
ReleasedFeb 2026
Price in / out$5.0 / $25
Context1M
Core tests taken7 of 9
163147–179
WeightsOpen
ReleasedSep 2026
Price in / out$0.051 / $0.57
Context1.05M
Core tests taken6 of 9
Claude Opus 4.6 and DeepSeek-V4.1-Flash are too close to call (2 points apart, within the uncertainty), ahead on 2 of 4 shared core tests, at 56× the price.
Test by testthird-party results
Claude Opus 4.6
DeepSeek-V4.1-Flash
Every result as a table
| Test | Claude Opus 4.6 | DeepSeek-V4.1-Flash |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 90.5% | 89.8% |
| LiveBench Data Analysis | 69.9% | 79.3% |
| LiveBench IF | 63.3% | 70.0% |
| LiveBench Language | 83.3% | 81.2% |
| LiveBench Reasoning | 88.7% | 86.7% |
| SimpleBench | 67.6% | 66.7% |
| SimpleQA Verified | 47.0% | 42.9% |
| Maths | ||
| FrontierMath (tiers 1-3) | 66.0% | 67.4% |
| AIME-style maths (OTIS mock) | 94.4% | 98.3% |
| LiveBench Mathematics | 89.3% | 93.3% |
| Chess puzzles | 21.0% | 48.0% |
| FrontierMath (tier 4) | 26.8% | 26.8% |
| Coding | ||
| LiveBench Agentic Coding | 49.0% | 77.3% |
| LiveBench Coding | 78.2% | 80.0% |
| Vibe Code Bench v1.1 | 57.6% | 84.7% |
| MirrorCode | 33.3% | 64.4% |
| Agents | ||
| APEX-Agents | 46.3% | 39.5% |
| DTBench | 91.2% | 89.9% |
| Furniture assembly | 28.3% | 34.2% |
| LMCA | 55.8% | 47.0% |
| Mystery game puzzles | 25.0% | 43.0% |
| Professional work | ||
| LegalBench | 85.3% | 83.3% |
| MedCode | 49.1% | 41.2% |
| MedScribe | 86.7% | 85.5% |
| Novel problems | ||
| ARC-AGI-2 | 69.2% | 72.9% |
| ARC-AGI-1 | 94.0% | 94.5% |
Task length · METR47 sec
Tests at least two of these models have taken. Bold rows are the core tests.