Up to four models
DeepSeek-V4-Pro vs DeepSeek-V4-Pro (Apr 2026)
On the index (GPT-4 = 100), DeepSeek-V4-Pro scores 169 and DeepSeek-V4-Pro (Apr 2026) 154. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
DeepSeek-V4-Pro169152–185
WeightsOpen
ReleasedAug 2026
Price in / out$1.32 / $3.96
Context1.05M
Core tests taken6 of 9
DeepSeek-V4-Pro (Apr 2026)154137–170
WeightsOpen
ReleasedApr 2026
Price in / out$0.96 / $1.91
Context1.05M
Core tests taken3 of 9
DeepSeek-V4-Pro leads DeepSeek-V4-Pro (Apr 2026) by 15 points, ahead on 2 of 2 shared core tests, at 1.7× the price.
Test by testthird-party results
DeepSeek-V4-ProDeepSeek-V4-Pro (Apr 2026)Every result as a table
| Test | DeepSeek-V4-Pro | DeepSeek-V4-Pro (Apr 2026) |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 91.7% | 90.9% |
| LiveBench Data Analysis | 79.2% | 74.5% |
| LiveBench IF | 67.7% | 62.4% |
| LiveBench Language | 82.1% | 78.1% |
| LiveBench Reasoning | 85.8% | 82.7% |
| SimpleQA Verified | 52.9% | 47.0% |
| Maths | ||
| FrontierMath (tiers 1-3) | 64.6% | 45.3% |
| AIME-style maths (OTIS mock) | 98.6% | 96.7% |
| LiveBench Mathematics | 95.1% | 90.7% |
| Chess puzzles | 46.0% | 19.0% |
| FrontierMath (tier 4) | 26.8% | 2.4% |
| Coding | ||
| LiveBench Agentic Coding | 54.9% | 42.6% |
| LiveBench Coding | 77.2% | 70.0% |
| WeirdML | 66.2% | 48.9% |
| Agents | ||
| DTBench | 93.9% | 90.7% |
| LMCA | 45.5% | 41.2% |
| Mystery game puzzles | 43.0% | 17.0% |
Tests at least two of these models have taken. Bold rows are the core tests.
