Up to four models
DeepSeek-V4-Pro vs Step 5 Preview
On the index (GPT-4 = 100), DeepSeek-V4-Pro scores 163 and Step 5 Preview 163. 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
163145–180
WeightsClosed
ReleasedSep 2026
Price in / out$1.0 / $2.7
Context1M
Core tests taken2 of 9
DeepSeek-V4-Pro and Step 5 Preview are too close to call (0 points apart, within the uncertainty), ahead on 0 of 2 shared core tests, at a similar price.
Test by testthird-party results
Every result as a table
| Test | DeepSeek-V4-Pro | Step 5 Preview |
|---|---|---|
| Maths | ||
| ProofBench v1.1 | 50.0% | 42.0% |
| Coding | ||
| Terminal-Bench 4.0 | 14.1% | 31.8% |
| Code Migration | 41.5% | 39.0% |
| IOI | 51.6% | 38.1% |
| Vibe Code Bench v1.1 | 82.3% | 70.4% |
| Agents | ||
| SRE Bench | 1.9% | 0.4% |
| Terminal-Bench-Science | 4.3% | 2.9% |
| Professional work | ||
| Finance Agent (v2) | 50.4% | 50.7% |
| EMB | 52.8% | 57.1% |
| HLAB-Vals-1.1 | 10.8% | 12.9% |
| Legal Research Bench | 40.9% | 40.4% |
| Tax Agent Bench | 26.8% | 27.3% |
Tests at least two of these models have taken. Bold rows are the core tests.