Up to four models
DeepSeek-V4-Pro (Apr 2026) vs o3
On the index (GPT-4 = 100), DeepSeek-V4-Pro (Apr 2026) scores 147 and o3 147. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
147131–164
WeightsOpen
ReleasedApr 2026
Price in / out$0.90 / $1.5
Context1.05M
Core tests taken3 of 9
DeepSeek-V4-Pro (Apr 2026) and o3 are too close to call (0 points apart, within the uncertainty), ahead on 3 of 3 shared core tests, and costs 3.3× less.
Test by testthird-party results
Every result as a table
| Test | DeepSeek-V4-Pro (Apr 2026) | o3 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 90.9% | 81.8% |
| SimpleQA Verified | 47.0% | 49.4% |
| Maths | ||
| FrontierMath (tiers 1-3) | 45.3% | 33.3% |
| AIME-style maths (OTIS mock) | 96.7% | 84.4% |
| Chess puzzles | 19.0% | 37.0% |
| Coding | ||
| SWE-bench Verified | 77.6% | 62.3% |
| WeirdML | 48.6% | 52.4% |
| Agents | ||
| DTBench | 90.7% | 84.8% |
| LMCA | 41.2% | 39.7% |
| Mystery game puzzles | 17.0% | 29.0% |
Task length · METR39 sec
Tests at least two of these models have taken. Bold rows are the core tests.