Up to four models
Step 5 Preview vs GPT-6 Luna
On the index (GPT-4 = 100), Step 5 Preview scores 163 and GPT-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.
163145–180
WeightsClosed
ReleasedSep 2026
Price in / out$1.0 / $2.7
Context1M
Core tests taken2 of 9
GPT-6 Luna
162146–178
WeightsClosed
ReleasedSep 2026
Price in / out$0.10 / $0.50
Context1.05M
Core tests taken6 of 9
Step 5 Preview and GPT-6 Luna are too close to call (0 points apart, within the uncertainty), ahead on 2 of 2 shared core tests, at 7.1× the price.
Test by testthird-party results
Every result as a table
| Test | Step 5 Preview | GPT-6 Luna |
|---|---|---|
| Maths | ||
| ProofBench v1.1 | 42.0% | 64.0% |
| Coding | ||
| Terminal-Bench 4.0 | 31.8% | 13.6% |
| Code Migration | 39.0% | 42.5% |
| IOI | 38.1% | 55.6% |
| Vibe Code Bench v1.1 | 70.4% | 81.7% |
| Agents | ||
| BioMysteryBench | 67.4% | 61.5% |
| SRE Bench | 0.4% | 2.7% |
| Terminal-Bench-Science | 2.9% | 4.3% |
| Professional work | ||
| Finance Agent (v2) | 50.7% | 49.9% |
| EMB | 57.1% | 68.5% |
| HLAB-Vals-1.1 | 12.9% | 4.2% |
| Legal Research Bench | 40.4% | 30.3% |
| Public Benefits Bench v1.1 | 58.1% | 57.6% |
| Tax Agent Bench | 27.3% | 20.0% |
| Novel problems | ||
| MysteryMechanism | 18.5% | 19.4% |
Tests at least two of these models have taken. Bold rows are the core tests.