Model comparison
GPT-5.4 vs GPT-6 Luna
GPT-5.4 is the stronger model overall, scoring 59.4 to 53.3 on the Noometry Index. GPT-6 Luna costs 28× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Last verified . 40 shared benchmarks.
Summary
- They share 40 benchmarks with published results for both. GPT-5.4 scores higher in 8 categories and GPT-6 Luna in 2 categories; 10 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where GPT-5.4 leads 71.9 to 58.3.
- The biggest single-benchmark swing is Mystery Game Puzzles: 37% for GPT-5.4 and 7% for GPT-6 Luna.
- GPT-6 Luna is cheaper at $0.10 / $0.50 per million input/output tokens, against $2.50 / $15 for GPT-5.4.
Side by side
| GPT-5.4 | GPT-6 Luna | |
|---|---|---|
| Provider | OpenAI | OpenAI |
| Noometry Index | 59.4 | 53.3 |
| Released | 2026-03-05 | 2026-09-22 |
| Weights | Proprietary | Proprietary |
| Context window | 1.05M | 1.05M |
| Max output | 128K | 128K |
| Input $ / M tokens | $2.50 | $0.10 |
| Output $ / M tokens | $15 | $0.50 |
| Results tracked | 68 | 42 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-6 Luna leads
GPT-5.4: 52.6 (#33), GPT-6 Luna: 55.5 (#25)
| Benchmark | GPT-5.4 | GPT-6 Luna |
|---|---|---|
| DeepSWE | 51.8% | 66.6% |
| LMArena WebDev | 1465 | 1581 |
| SciCode | 56.6% | 54.6% |
| LMArena Coding | 1497 | 1439 |
| ALE-Bench | 1,607 | 1,577 |
| SWE-bench Verified | 76.9% | — |
| FrontierCode | — | 42.4% |
| GSO | 31.4% | — |
| WeirdML | 77.7% | — |
| MirrorCode | 15.6% | — |
| AlgoTune | 1.85 | — |
Agentic & Tool Use GPT-5.4 leads
GPT-5.4: 46.5 (#13), GPT-6 Luna: 33.3 (#54)
| Benchmark | GPT-5.4 | GPT-6 Luna |
|---|---|---|
| APEX-Agents | 52.4% | 44.3% |
| Terminal-Bench | 81.8% | — |
| τ²-bench Banking | 39.4% | — |
| DeepResearch Bench | 35.1% | — |
| PostTrainBench | 19% | — |
| GBAEval | 45.1% | — |
| GDP.pdf | — | 23% |
| LMArena Search | 1197 | — |
| METR Time Horizons | 74.3% | — |
| Vending-Bench 2 | 6,144 | — |
Reasoning GPT-5.4 leads
GPT-5.4: 61.8 (#19), GPT-6 Luna: 48.2 (#41)
| Benchmark | GPT-5.4 | GPT-6 Luna |
|---|---|---|
| ARC-AGI-2 | 74% | 59.3% |
| NYT Connections (extended) | 91.3% | 68.7% |
| ARC-AGI-1 | 93.7% | 86.7% |
| CritPt | 23.4% | 19.4% |
| Chess Puzzles | 44% | 31% |
| LMArena Hard Prompts | 1485 | 1411 |
| Mystery Game Puzzles | 37% | 7% |
| DTBench | 94.4% | 90.1% |
| LMCA | 52% | 44.5% |
| Epoch Capabilities Index | 156.81 | 156.28 |
| Kagi LLM Benchmark | 63.8% | — |
| EnigmaEval | 16% | — |
| Thematic Generalization | 80% | — |
| EBR-Bench | 25.4% | — |
| ForecastBench | 59.5 | — |
Math GPT-6 Luna leads
GPT-5.4: 73.5 (#19), GPT-6 Luna: 76.1 (#15)
| Benchmark | GPT-5.4 | GPT-6 Luna |
|---|---|---|
| FrontierMath (Tiers 1-3) | 78.6% | 78.9% |
| FrontierMath Tier 4 | 49% | 56.1% |
| OTIS Mock AIME 2024-2025 | 97.8% | 98.9% |
| ProofBench | 56% | 64% |
| LMArena Math | 1488 | 1416 |
| MathArena Final-Answer Competitions | 83.1% | — |
| FrontierMath (Feb 2025 set) | 47.6% | — |
| FrontierMath Tier 4 (v1) | 27.1% | — |
Knowledge GPT-5.4 leads
GPT-5.4: 65.3 (#14), GPT-6 Luna: 57.0 (#41)
| Benchmark | GPT-5.4 | GPT-6 Luna |
|---|---|---|
| GPQA Diamond | 93.3% | 90.5% |
| SimpleQA Verified | 45.1% | 41.4% |
| LMArena Expert | 1507 | 1444 |
| Humanity's Last Exam | 36.2% | — |
| Vectara Hallucination Rate | 7% | — |
Multimodal GPT-5.4 leads
GPT-5.4: 43.7 (#20), GPT-6 Luna: 42.4 (#30)
| Benchmark | GPT-5.4 | GPT-6 Luna |
|---|---|---|
| LMArena Vision | 1303 | 1217 |
| Blueprint-Bench 2 | 27.1% | 31.2% |
| Furniture Assembly | 37.5% | 44.2% |
| LMArena Document | 1471 | — |
Multilingual GPT-5.4 leads
GPT-5.4: 56.2 (#23), GPT-6 Luna: 50.5 (#117)
| Benchmark | GPT-5.4 | GPT-6 Luna |
|---|---|---|
| LMArena Non-English | 1465 | 1386 |
| LMArena Chinese | 1519 | 1433 |
| LMArena French | 1493 | 1420 |
| LMArena German | 1472 | 1369 |
| LMArena Japanese | 1485 | 1369 |
| LMArena Korean | 1448 | 1360 |
| LMArena Russian | 1480 | 1394 |
| LMArena Spanish | 1454 | 1393 |
Instruction Following GPT-5.4 leads
GPT-5.4: 77.1 (#27), GPT-6 Luna: 74.3 (#99)
| Benchmark | GPT-5.4 | GPT-6 Luna |
|---|---|---|
| LMArena Instruction Following | 1469 | 1409 |
Long Context GPT-5.4 leads
GPT-5.4: 50.3 (#8), GPT-6 Luna: 43.0 (#111)
| Benchmark | GPT-5.4 | GPT-6 Luna |
|---|---|---|
| LMArena Longer Query | 1473 | 1409 |
| CL-bench | 27.9% | — |
| CL-bench Life | 21.7% | — |
Writing & Preference GPT-5.4 leads
GPT-5.4: 71.9 (#17), GPT-6 Luna: 58.3 (#119)
| Benchmark | GPT-5.4 | GPT-6 Luna |
|---|---|---|
| LMArena Text | 1469 | 1391 |
| LMArena Creative Writing | 1439 | 1363 |
| LMArena Multi-Turn | 1482 | 1396 |
| EQ-Bench Creative Writing | 1840 | — |
| EQ-Bench 4 | 1272 | — |
Frequently asked questions
Is GPT-5.4 better than GPT-6 Luna?
GPT-5.4 is the stronger model overall, scoring 59.4 to 53.3 on the Noometry Index. GPT-6 Luna costs 28× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Which is cheaper, GPT-5.4 or GPT-6 Luna?
GPT-6 Luna is cheaper. It lists at $0.10 per million input tokens and $0.50 per million output tokens; GPT-5.4 lists at $2.50 and $15.
Is GPT-5.4 or GPT-6 Luna better for coding?
GPT-6 Luna scores higher on coding benchmarks: 55.5 versus 52.6 in the Noometry coding category.
Which has the bigger context window?
Both accept 1.05M tokens.
How many benchmarks do GPT-5.4 and GPT-6 Luna share?
40 benchmarks have published results for both models. GPT-5.4 has 68 scored results on Noometry and GPT-6 Luna has 42.