GPT-5.4 vs GPT-5.6 Luna
Updated October 21, 2026. 87 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick GPT-5.6 Luna for coding: it scores 61.5 on the Vibe Coding Index against 37.7 for GPT-5.4.
Pick GPT-5.4 when what matters is ProgramBench (Vals AI) or WeirdML, where it leads.
- Shared results
- 87
- GPT-5.4 only
- 22
- GPT-5.6 Luna only
- 29
- Like-for-like categories
- 8 of 9
Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, four public coding benchmarks
GPT-5.6 Luna
GPT-5.6 Luna scores 61.5 against 37.7.
1 and 2 benchmarks behind the scores
Intelligence
AA Intelligence Index
GPT-5.4
GPT-5.4 leads, 39.0 against 37.3.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only GPT-5.6 Luna has a AA Agentic Index result, 42.1.
Directional only
Reasoning
GPQA Diamond
No call
Tied at 91.7%.
Same board, both measured
Math
AIME
No call
Only GPT-5.4 has a AIME result, 96.7%.
Directional only
Knowledge
MMLU Pro
GPT-5.4
GPT-5.4 leads, 87.5% against 86.0%.
Same board, both measured
Web design
Design Arena: websites
GPT-5.6 Luna
GPT-5.6 Luna leads, 1256 against 1199.
Same board, both measured
Long documents
Listed context window
No call
Both take 1.05M tokens.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
GPT-5.6 Luna
GPT-5.4 $0.010, GPT-5.6 Luna $0.0008 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
GPT-5.6 Luna
GPT-5.4 $1.02, GPT-5.6 Luna $0.04.
Published attempt cost and pass rate
At a glance
- Coding index
- GPT-5.4 37.7 from 1 benchmark, GPT-5.6 Luna 61.5 from 2 benchmarks.
- Widest category
- Agentic work, 11.1 points apart on average over 2 shared benchmarks.
- Widest single gap
- WeirdML: GPT-5.4 77.7% against GPT-5.6 Luna 60.9%, 16.8 points apart on a 0 to 100 scale.
- Output price
- GPT-5.4 $15.00 against GPT-5.6 Luna $1.20 per million output tokens, 12.5x apart.
- Context window
- Both take 1.05M tokens.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | GPT-5.4 | Bars | GPT-5.6 Luna | Ties |
|---|---|---|---|---|
| Coding | 4 | 10 | ||
| Intelligence | 2 | 0 | ||
| Agentic work | 2 | 0 | ||
| Reasoning | 13 | 0 | 1 tie | |
| Math | 2 | 4 | ||
| Knowledge | 9 | 2 | ||
| Web design | 1 | 7 | ||
| General | 26 | 4 | ||
| All shared boards | 59 | 27 | 1 tie |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index410
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Leader |
|---|---|---|---|
| LiveBench coding | Not measured | 73.6% | Only GPT-5.6 Luna measured |
| LMArena WebDev | 1464 | 1518 | GPT-5.6 Luna |
| AA Coding Index | 71.1 | 71.4 | GPT-5.6 Luna |
| LiveCodeBench (Vals AI) | 84.1% | Not measured | Only GPT-5.4 measured |
| SWE-bench Verified (Vals AI) | 78.2% | 93.0% | GPT-5.6 Luna |
| Terminal-Bench 4 (Vals AI) | Not measured | 11.6% | Only GPT-5.6 Luna measured |
| Vibe Code Bench (Vals AI) | 67.4% | 77.1% | GPT-5.6 Luna |
| Code migration (Vals AI) | 35.0% | 44.5% | GPT-5.6 Luna |
| ProgramBench (Vals AI) | 0.5% | 0.0% | GPT-5.4 |
| IOI (Vals AI) | Not measured | 61.8% | Only GPT-5.6 Luna measured |
| Terminal-Bench 2.1 (Vals AI) | Not measured | 79.0% | Only GPT-5.6 Luna measured |
| Vibe Code Bench 1-100 (Vals AI) | Not measured | 22.6% | Only GPT-5.6 Luna measured |
| SkillsBench (Vals AI) | 51.7% | 60.5% | GPT-5.6 Luna |
| CyberBench (Vals AI) | Not measured | 73.6% | Only GPT-5.6 Luna measured |
| SciCode | Not measured | 53.6% | Only GPT-5.6 Luna measured |
| Terminal-Bench Hard (Artificial Analysis run) | 57.6% | Not measured | Only GPT-5.4 measured |
| Terminal-Bench | 81.8% | Not measured | Only GPT-5.4 measured |
| FrontierCode | Not measured | 39.8% | Only GPT-5.6 Luna measured |
| GSO-Bench | 31.4% | Not measured | Only GPT-5.4 measured |
| WeirdML | 77.7% | 60.9% | GPT-5.4 |
| MirrorCode | 15.6% | Not measured | Only GPT-5.4 measured |
| ALE-Bench | 1607.0 | 1667.4 | GPT-5.6 Luna |
| LMArena Text: coding | 1496 | 1466 | GPT-5.4 |
| LMArena Text: software and IT | 1491 | 1463 | GPT-5.4 |
| LMArena WebDev: React | 1454 | 1508 | GPT-5.6 Luna |
| LMArena WebDev: HTML | 1478 | 1520 | GPT-5.6 Luna |
| LMArena WebDev: from an image | 1436 | 1517 | GPT-5.6 Luna |
| Neither model has a result on the other 11 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Leader |
|---|---|---|---|
| AA Intelligence Index | 39.0 | 37.3 | GPT-5.4 |
| Epoch Capabilities Index | 157.3 | 156.5 | GPT-5.4 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 42.1 | Only GPT-5.6 Luna measured |
| Design Arena: agents building web apps | 1059 | Not measured | Only GPT-5.4 measured |
| Design Arena: agents building full-stack apps | 1017 | Not measured | Only GPT-5.4 measured |
| Design Arena: agents building mobile apps | 1087 | Not measured | Only GPT-5.4 measured |
| Design Arena: agents building Android apps | 1037 | Not measured | Only GPT-5.4 measured |
| Design Arena: agents building Godot games | 1135 | Not measured | Only GPT-5.4 measured |
| LiveBench research | Not measured | 51.0% | Only GPT-5.6 Luna measured |
| LiveBench browser | Not measured | 68.0% | Only GPT-5.6 Luna measured |
| LiveBench long agentic | Not measured | 30.4% | Only GPT-5.6 Luna measured |
| SREBench (Vals AI) | Not measured | 1.5% | Only GPT-5.6 Luna measured |
| Finance Agent (Vals AI) | Not measured | 55.0% | Only GPT-5.6 Luna measured |
| Tax Agent Bench (Vals AI) | Not measured | 24.8% | Only GPT-5.6 Luna measured |
| Legal Research Bench (Vals AI) | Not measured | 36.5% | Only GPT-5.6 Luna measured |
| Excel Modeling Benchmark (Vals AI) | Not measured | 67.1% | Only GPT-5.6 Luna measured |
| BioMysteryBench (Vals AI) | Not measured | 61.5% | Only GPT-5.6 Luna measured |
| Terminal-Bench-Science (Vals AI) | Not measured | 0.0% | Only GPT-5.6 Luna measured |
| Vals Science Index | Not measured | 25.3% | Only GPT-5.6 Luna measured |
| τ²-Bench Telecom | 87.1% | Not measured | Only GPT-5.4 measured |
| APEX-Agents | 52.4% | 43.0% | GPT-5.4 |
| METR time horizons | 74.3% | Not measured | Only GPT-5.4 measured |
| BALROG | Not measured | 45.6% | Only GPT-5.6 Luna measured |
| DeepResearch Bench | 35.1% | Not measured | Only GPT-5.4 measured |
| PostTrainBench | 19.0% | Not measured | Only GPT-5.4 measured |
| Vending-Bench 2 | $6,144 | $4,095 | GPT-5.4 |
| Neither model has a result on the other 15 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index130
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Leader |
|---|---|---|---|
| GPQA Diamond | 91.7% | 91.7% | Tie |
| LiveBench reasoning | Not measured | 85.6% | Only GPT-5.6 Luna measured |
| MysteryMechanism (Vals AI) | Not measured | 14.4% | Only GPT-5.6 Luna measured |
| MMMU Pro (Vals AI) | 87.5% | 85.0% | GPT-5.4 |
| GPQA Diamond (Artificial Analysis run) | 92.0% | 91.1% | GPT-5.4 |
| Humanity's Last Exam (Artificial Analysis run) | 43.7% | 39.5% | GPT-5.4 |
| GPQA Diamond (Epoch AI run) | 93.3% | 91.6% | GPT-5.4 |
| ARC-AGI | 93.7% | 88.0% | GPT-5.4 |
| ARC-AGI-2 | 74.0% | 59.5% | GPT-5.4 |
| SimpleBench | Not measured | 46.8% | Only GPT-5.6 Luna measured |
| Chess puzzles | 48.0% | 43.0% | GPT-5.4 |
| Mystery game puzzles | 37.0% | 21.0% | GPT-5.4 |
| EBR-bench | 25.4% | Not measured | Only GPT-5.4 measured |
| LMArena Text: hard prompts | 1485 | 1451 | GPT-5.4 |
| LMArena Text: expert | 1507 | 1478 | GPT-5.4 |
| LMArena Text: hard prompts in English | 1479 | 1460 | GPT-5.4 |
| LMArena Vision: diagrams | 1327 | 1273 | GPT-5.4 |
| LMArena Vision: homework | 1336 | 1282 | GPT-5.4 |
| Neither model has a result on the other 3 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index24
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Leader |
|---|---|---|---|
| AIME | 96.7% | Not measured | Only GPT-5.4 measured |
| OTIS Mock AIME | 97.8% | 98.3% | GPT-5.6 Luna |
| FrontierMath, tiers 1 to 3 | 78.6% | 82.1% | GPT-5.6 Luna |
| FrontierMath, tier 4 | 49.0% | 61.0% | GPT-5.6 Luna |
| ProofBench | 56.0% | 60.0% | GPT-5.6 Luna |
| LMArena Text: math | 1488 | 1459 | GPT-5.4 |
| LMArena Text: mathematical work | 1497 | 1470 | GPT-5.4 |
| Neither model has a result on the other 6 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index92
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Leader |
|---|---|---|---|
| MMLU Pro | 87.5% | 86.0% | GPT-5.4 |
| LegalBench (Vals AI) | 86.0% | 84.0% | GPT-5.4 |
| CaseLaw (Vals AI) | 63.8% | Not measured | Only GPT-5.4 measured |
| CorpFin (Vals AI) | 65.3% | 64.2% | GPT-5.4 |
| TaxEval (Vals AI) | 74.0% | 76.2% | GPT-5.6 Luna |
| MortgageTax (Vals AI) | 68.3% | 67.3% | GPT-5.4 |
| MedQA (Vals AI) | 96.1% | Not measured | Only GPT-5.4 measured |
| MedCode (Vals AI) | 41.3% | 42.4% | GPT-5.6 Luna |
| SimpleQA Verified | 45.1% | 41.0% | GPT-5.4 |
| LMArena Text: science | 1479 | 1433 | GPT-5.4 |
| LMArena Text: medicine and healthcare | 1458 | 1399 | GPT-5.4 |
| LMArena Text: legal and government | 1486 | 1446 | GPT-5.4 |
| LMArena Text: business and finance | 1474 | 1439 | GPT-5.4 |
| Neither model has a result on the other 4 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index17
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Leader |
|---|---|---|---|
| Design Arena: websites | 1199 | 1256 | GPT-5.6 Luna |
| Design Arena: UI components | 1215 | 1250 | GPT-5.6 Luna |
| Design Arena: game dev | 1195 | 1196 | GPT-5.6 Luna |
| Design Arena: data viz | 1196 | 1212 | GPT-5.6 Luna |
| Design Arena: 3D | 1065 | 1121 | GPT-5.6 Luna |
| Design Arena: all code categories | 1180 | 1231 | GPT-5.6 Luna |
| Design Arena: SVG | 1141 | 1193 | GPT-5.6 Luna |
| Design Arena: ASCII art | 1188 | 1144 | GPT-5.4 |
| LiveBench design | Not measured | 76.0% | Only GPT-5.6 Luna measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index264
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Leader |
|---|---|---|---|
| LiveBench data analysis | Not measured | 78.0% | Only GPT-5.6 Luna measured |
| Public Benefits Bench (Vals AI) | Not measured | 61.2% | Only GPT-5.6 Luna measured |
| SAGE (Vals AI) | 43.3% | 44.2% | GPT-5.6 Luna |
| MedScribe (Vals AI) | 77.5% | 84.4% | GPT-5.6 Luna |
| Vals Multimodal Index | Not measured | 69.1% | Only GPT-5.6 Luna measured |
| IFBench | 73.9% | Not measured | Only GPT-5.4 measured |
| Long context reasoning (AA-LCR) | 82.0% | 83.7% | GPT-5.6 Luna |
| DTBench | 94.4% | 89.1% | GPT-5.4 |
| LMCA | 52.0% | 48.5% | GPT-5.4 |
| CL-bench | 27.9% | Not measured | Only GPT-5.4 measured |
| Surface Evolver Bench | Not measured | 61.9% | Only GPT-5.6 Luna measured |
| Furniture assembly | 37.5% | 42.5% | GPT-5.6 Luna |
| CL-bench Life | 21.7% | Not measured | Only GPT-5.4 measured |
| LMArena Text | 1469 | 1431 | GPT-5.4 |
| LMArena Text: instruction following | 1469 | 1437 | GPT-5.4 |
| LMArena Text: creative writing | 1439 | 1396 | GPT-5.4 |
| LMArena Vision | 1303 | 1257 | GPT-5.4 |
| LMArena Text: English | 1467 | 1441 | GPT-5.4 |
| LMArena Text: non-English | 1465 | 1417 | GPT-5.4 |
| LMArena Text: Chinese | 1520 | 1470 | GPT-5.4 |
| LMArena Text: German | 1472 | 1455 | GPT-5.4 |
| LMArena Text: Japanese | 1485 | 1415 | GPT-5.4 |
| LMArena Text: Korean | 1449 | 1415 | GPT-5.4 |
| LMArena Text: French | 1493 | 1458 | GPT-5.4 |
| LMArena Text: Spanish | 1452 | 1446 | GPT-5.4 |
| LMArena Text: Russian | 1480 | 1427 | GPT-5.4 |
| LMArena Text: Polish | 1472 | 1405 | GPT-5.4 |
| LMArena Text: multi-turn | 1482 | 1434 | GPT-5.4 |
| LMArena Text: longer queries | 1473 | 1436 | GPT-5.4 |
| LMArena Text: writing and language | 1458 | 1419 | GPT-5.4 |
| LMArena Text: entertainment and media | 1438 | 1402 | GPT-5.4 |
| LMArena Vision: English | 1301 | 1263 | GPT-5.4 |
| LMArena Vision: Chinese | 1354 | 1294 | GPT-5.4 |
| LMArena Vision: reading text | 1312 | 1264 | GPT-5.4 |
| LMArena Vision: humor | 1284 | 1190 | GPT-5.4 |
| LMArena Vision: recognizing things | 1251 | Not measured | Only GPT-5.4 measured |
| LMArena Vision: creative writing | 1283 | 1257 | GPT-5.4 |
| LMArena Document | 1471 | 1457 | GPT-5.4 |
| Neither model has a result on the other 8 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
GPT-5.6 Luna costs less
Repository review
50K input + 3K output tokens
GPT-5.6 Luna costs less
Agent loop
200K input + 10K output tokens
GPT-5.6 Luna costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | OpenAI GPT-5.4 | OpenAI GPT-5.6 Luna |
|---|---|---|
| Maker | OpenAI | OpenAI |
| Weights | No public weights listed | No public weights listed |
| Released | March 5, 2026 | July 9, 2026 |
| Context window | 1.05M | 1.05M |
| Input $/Mtok | $2.50 | $0.20 |
| Output $/Mtok | $15.00 | $1.20 |
| Coding index | 37.7 | 61.5 |
| Independent measurements behind it | 1 | 2 |
| Cost per solved task | $1.02 | $0.04 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-5.4 better than GPT-5.6 Luna for coding?
Pick GPT-5.6 Luna for coding: it scores 61.5 on the Vibe Coding Index against 37.7 for GPT-5.4. Pick GPT-5.4 when what matters is ProgramBench (Vals AI) or WeirdML, where it leads.
Which is cheaper, GPT-5.4 or GPT-5.6 Luna?
GPT-5.4 costs $2.50 per million input tokens and $15.00 per million output tokens; GPT-5.6 Luna costs $0.20 and $1.20. Per solved SWE-bench Verified task, GPT-5.4 costs $1.02 and GPT-5.6 Luna $0.04.
Which has the bigger context window, GPT-5.4 or GPT-5.6 Luna?
Neither: both take 1.05M tokens.
How many benchmarks is the GPT-5.4 vs GPT-5.6 Luna comparison based on?
87 benchmarks have a result for both models as of October 21, 2026. GPT-5.4's coding index rests on 1 independent benchmark and GPT-5.6 Luna's on 2. Where only one model has a result, the page says so instead of naming a winner.