Model comparison
GPT-6 Luna vs Step 5 Preview
GPT-6 Luna is the stronger model overall, scoring 53.3 to 47.9 on the Noometry Index.
Last verified . 18 shared benchmarks.
StepFunSummary
- They share 18 benchmarks with published results for both. GPT-6 Luna scores higher in 5 categories and Step 5 Preview in 4 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where GPT-6 Luna leads 76.1 to 46.1.
- The biggest single-benchmark swing is ProofBench: 64% for GPT-6 Luna and 42% for Step 5 Preview.
- GPT-6 Luna is cheaper at $0.10 / $0.50 per million input/output tokens, against $1 / $2.70 for Step 5 Preview.
- GPT-6 Luna accepts more context: 1.05M tokens versus 1.02M.
Side by side
| GPT-6 Luna | Step 5 Preview | |
|---|---|---|
| Provider | OpenAI | StepFun |
| Noometry Index | 53.3 | 47.9 |
| Released | 2026-09-22 | 2026-09-16 |
| Weights | Proprietary | Proprietary |
| Context window | 1.05M | 1.02M |
| Max output | 128K | 66K |
| Input $ / M tokens | $0.10 | $1 |
| Output $ / M tokens | $0.50 | $2.70 |
| Results tracked | 42 | 18 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-6 Luna leads
GPT-6 Luna: 55.5 (#25), Step 5 Preview: 51.8 (#36)
| Benchmark | GPT-6 Luna | Step 5 Preview |
|---|---|---|
| LMArena WebDev | 1581 | 1564 |
| SciCode | 54.6% | 58.9% |
| LMArena Coding | 1439 | 1480 |
| DeepSWE | 66.6% | — |
| FrontierCode | 42.4% | — |
| ALE-Bench | 1,577 | — |
Agentic & Tool Use Not comparable
GPT-6 Luna: 33.3 (#54), Step 5 Preview: —
| Benchmark | GPT-6 Luna | Step 5 Preview |
|---|---|---|
| APEX-Agents | 44.3% | — |
| GDP.pdf | 23% | — |
Reasoning GPT-6 Luna leads
GPT-6 Luna: 48.2 (#41), Step 5 Preview: 40.0 (#57)
| Benchmark | GPT-6 Luna | Step 5 Preview |
|---|---|---|
| CritPt | 19.4% | 20.9% |
| LMArena Hard Prompts | 1411 | 1465 |
| ARC-AGI-2 | 59.3% | — |
| NYT Connections (extended) | 68.7% | — |
| ARC-AGI-1 | 86.7% | — |
| Chess Puzzles | 31% | — |
| Mystery Game Puzzles | 7% | — |
| DTBench | 90.1% | — |
| LMCA | 44.5% | — |
| Epoch Capabilities Index | 156.28 | — |
Math GPT-6 Luna leads
GPT-6 Luna: 76.1 (#15), Step 5 Preview: 46.1 (#72)
| Benchmark | GPT-6 Luna | Step 5 Preview |
|---|---|---|
| ProofBench | 64% | 42% |
| LMArena Math | 1416 | 1470 |
| FrontierMath (Tiers 1-3) | 78.9% | — |
| FrontierMath Tier 4 | 56.1% | — |
| OTIS Mock AIME 2024-2025 | 98.9% | — |
Knowledge GPT-6 Luna leads
GPT-6 Luna: 57.0 (#41), Step 5 Preview: 41.2 (#112)
| Benchmark | GPT-6 Luna | Step 5 Preview |
|---|---|---|
| LMArena Expert | 1444 | 1470 |
| GPQA Diamond | 90.5% | — |
| SimpleQA Verified | 41.4% | — |
Multimodal GPT-6 Luna leads
GPT-6 Luna: 42.4 (#30), Step 5 Preview: 41.0 (#41)
| Benchmark | GPT-6 Luna | Step 5 Preview |
|---|---|---|
| LMArena Vision | 1217 | 1267 |
| Blueprint-Bench 2 | 31.2% | — |
| Furniture Assembly | 44.2% | — |
Multilingual Step 5 Preview leads
GPT-6 Luna: 50.5 (#117), Step 5 Preview: 53.8 (#54)
| Benchmark | GPT-6 Luna | Step 5 Preview |
|---|---|---|
| LMArena Non-English | 1386 | 1432 |
| LMArena Chinese | 1433 | 1519 |
| LMArena Russian | 1394 | 1436 |
| LMArena Spanish | 1393 | 1439 |
| LMArena French | 1420 | — |
| LMArena German | 1369 | — |
| LMArena Japanese | 1369 | — |
| LMArena Korean | 1360 | — |
Instruction Following Step 5 Preview leads
GPT-6 Luna: 74.3 (#99), Step 5 Preview: 76.0 (#50)
| Benchmark | GPT-6 Luna | Step 5 Preview |
|---|---|---|
| LMArena Instruction Following | 1409 | 1444 |
Long Context Step 5 Preview leads
GPT-6 Luna: 43.0 (#111), Step 5 Preview: 44.5 (#64)
| Benchmark | GPT-6 Luna | Step 5 Preview |
|---|---|---|
| LMArena Longer Query | 1409 | 1455 |
Writing & Preference Step 5 Preview leads
GPT-6 Luna: 58.3 (#119), Step 5 Preview: 62.8 (#71)
| Benchmark | GPT-6 Luna | Step 5 Preview |
|---|---|---|
| LMArena Text | 1391 | 1442 |
| LMArena Creative Writing | 1363 | 1410 |
| LMArena Multi-Turn | 1396 | 1447 |
Frequently asked questions
Is GPT-6 Luna better than Step 5 Preview?
GPT-6 Luna is the stronger model overall, scoring 53.3 to 47.9 on the Noometry Index.
Which is cheaper, GPT-6 Luna or Step 5 Preview?
GPT-6 Luna is cheaper. It lists at $0.10 per million input tokens and $0.50 per million output tokens; Step 5 Preview lists at $1 and $2.70.
Is GPT-6 Luna or Step 5 Preview better for coding?
GPT-6 Luna scores higher on coding benchmarks: 55.5 versus 51.8 in the Noometry coding category.
Which has the bigger context window?
GPT-6 Luna does, with 1.05M tokens against 1.02M.
How many benchmarks do GPT-6 Luna and Step 5 Preview share?
18 benchmarks have published results for both models. GPT-6 Luna has 42 scored results on Noometry and Step 5 Preview has 18.