Model comparison
GPT-5.4 vs GPT-5.6 Luna
Head-to-head evidence from 29 shared benchmark results across 6 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.
Public leaderboard positions: GPT-5.4 #8 (Supported); GPT-5.6 Luna #22 (Estimated). Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.
Evidence parity. GPT-5.4 and GPT-5.6 Luna share 29 comparable benchmark results. 5 of 8 categories are comparable. 23 results are unique to GPT-5.4; 12 to GPT-5.6 Luna.
Updated July 23, 2026- Shared results
- 29
- GPT-5.4 only
- 23
- GPT-5.6 Luna only
- 12
- Comparable categories
- 5 / 8
Pick GPT-5.4 if you want the stronger benchmark profile. GPT-5.6 Luna only becomes the better choice if knowledge is the priority or you want the cheaper token bill.
Confidence note. This is a partial-evidence comparison with 29 shared benchmark results across 6 evidence categories; 5 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.
Why this result
GPT-5.4 is clearly ahead on the BenchAlign aggregate, 74.24 to 67.17. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
GPT-5.4 is also the more expensive model on tokens at $2.50 input / $15.00 output per 1M tokens, versus $1.00 input / $6.00 output per 1M tokens for GPT-5.6 Luna. That is roughly 2.5x on output cost alone. GPT-5.4 gives you the larger context window at 1.05M, compared with 1M for GPT-5.6 Luna.
Category breakdown
Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.
| Category | GPT-5.4 | Δ | GPT-5.6 Luna |
|---|---|---|---|
| Knowledge | GPT-5.457.6 | Margin→ 34.7 | GPT-5.6 Luna92.3 |
| Math | GPT-5.442.5 | Margin→ 31.1 | GPT-5.6 Luna73.6 |
| Agentic | GPT-5.477.2 | Margin→ 6.9 | GPT-5.6 Luna84.1 |
| Multimodal | GPT-5.473.2 | Margin→ 5.2 | GPT-5.6 Luna78.4 |
| Coding | GPT-5.457.7 | Margin→ 5.0 | GPT-5.6 Luna62.7 |
Decisive benchmark drivers
The largest measured benchmark gaps in this matchup, with exact reported values.
More
- Source ↗
FrontierMath v2 (Tier 4)
MathA 27.100%B 58.500%Winner: GPT-5.6 LunaΔ 31.4FrontierMath v2 (Tier 4): GPT-5.4 scored 27.100%; GPT-5.6 Luna scored 58.500%. GPT-5.6 Luna wins this benchmark. - Source ↗
FrontierMath v2 (Tiers 1-3)
MathA 47.600%B 78.600%Winner: GPT-5.6 LunaΔ 31FrontierMath v2 (Tiers 1-3): GPT-5.4 scored 47.600%; GPT-5.6 Luna scored 78.600%. GPT-5.6 Luna wins this benchmark. - Source ↗
Terminal-Bench 2.0
AgenticA 75.1%B 84.7%Winner: GPT-5.6 LunaΔ 9.6Terminal-Bench 2.0: GPT-5.4 scored 75.1%; GPT-5.6 Luna scored 84.7%. GPT-5.6 Luna wins this benchmark. - Source ↗
SWE-bench Pro
CodingA 57.7%B 62.7%Winner: GPT-5.6 LunaΔ 5SWE-bench Pro: GPT-5.4 scored 57.7%; GPT-5.6 Luna scored 62.7%. GPT-5.6 Luna wins this benchmark. - Source ↗
MMMU-Pro
MultimodalA 81.2%B 78.4%Winner: GPT-5.4Δ 2.8MMMU-Pro: GPT-5.4 scored 81.2%; GPT-5.6 Luna scored 78.4%. GPT-5.4 wins this benchmark.
Operational comparison
Runtime and commercial metrics are compared only when both models have a complete sourced value.
| Metric | GPT-5.4 | GPT-5.6 Luna | Comparison |
|---|---|---|---|
| Input / output priceUSD per 1M tokens | GPT-5.4$2.5 input / $15 output | GPT-5.6 Luna$1 input / $6 output | GPT-5.6 Luna has the lower combined listed price. |
| Generation speedtokens per second | GPT-5.474 tok/s | GPT-5.6 LunaNot available | A complete speed comparison is not available. |
| First-answer latencyseconds to first token | GPT-5.4151.79 s | GPT-5.6 LunaNot available | A complete latency comparison is not available. |
| Context windowmaximum listed tokens | GPT-5.41.05M | GPT-5.6 Luna1M | GPT-5.4 lists the larger context window. |
Benchmark Deep Dive
AgenticGPT-5.6 Luna wins23 benchmarks
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Result |
|---|---|---|---|
| Terminal-Bench 2.0Source | 75.1% | 84.7% | GPT-5.6 Luna leads |
| CyberGymSource | 79.0% | 77.9% | GPT-5.4 leads |
| BrowseCompSource | 82.7% | 83.3% | GPT-5.6 Luna leads |
| OSWorld-VerifiedSource | 75% | — | Not comparable |
| MCP AtlasSource | 70.6% | — | Not comparable |
| ToolathlonSource | 54.6% | 53.4% | GPT-5.4 leads |
| τ²-bench resultsSource | 87.1% | — | Not comparable |
| Claw-EvalSource | 60.3% | — | Not comparable |
| DeepSearchQASource | 73.6% | — | Not comparable |
| AA Agentic IndexSource | 41.1% | 45.6% | GPT-5.6 Luna leads |
| APEX-Agents-AASource | 33.3% | 35.8% | GPT-5.6 Luna leads |
| GDPval-AASource | 44.7% | 54.2% | GPT-5.6 Luna leads |
| GDPval-AASource | 1395 | 1584 | GPT-5.6 Luna leads |
| Gert LabsSource | 64.89% | — | Not comparable |
| ResearchClawBenchSource | 15.3% | — | Not comparable |
| JobBenchSource | 38.9% | — | Not comparable |
| ExploitGymSource | 6.0% | 12.4% | GPT-5.6 Luna leads |
| OSWorld 2.0Source | — | 45.6% | Not comparable |
| AA Harvey LABSource | — | 87.9% | Not comparable |
| AA ITBenchSource | — | 40.3% | Not comparable |
| AA Tau3 BankingSource | — | 27.2% | Not comparable |
| AA AutomationBenchSource | — | 42.2% | Not comparable |
| aaTerminalBench21Source | — | 80.9% | Not comparable |
CodingGPT-5.6 Luna wins10 benchmarks
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Result |
|---|---|---|---|
| LiveCodeBench ProSource | 87.5% | — | Not comparable |
| SWE-bench ProSource | 57.7% | 62.7% | GPT-5.6 Luna leads |
| React Native EvalsSource | 85.3% | — | Not comparable |
| Vibe Code BenchSource | 67.42% | — | Not comparable |
| AA Coding IndexSource | 71.0% | 71.5% | GPT-5.6 Luna leads |
| AA-SciCodeSource | 56.6% | 52.5% | GPT-5.4 leads |
| Terminal-Bench 2.0Source | — | 84.7% | Not comparable |
| deepSweSource | — | 67.2% | Not comparable |
| FrontierCode 1.1 ExtendedSource | — | 55.1% | Not comparable |
| cursorBench32Source | — | 61.1% | Not comparable |
Reasoning3 benchmarks
KnowledgeGPT-5.6 Luna wins13 benchmarks
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Result |
|---|---|---|---|
| GPQASource | 92.8% | 92.3% | GPT-5.4 leads |
| HLESource | 52.1% | — | Not comparable |
| HLE w/o toolsSource | 39.8% | — | Not comparable |
| GPQA-DSource | 92.8% | 92.3% | GPT-5.4 leads |
| HealthBench HardSource | 40.1% | 32.0% | GPT-5.4 leads |
| MedXpertQA (Text)Source | 59.6% | — | Not comparable |
| Artificial Analysis Intelligence IndexSource | 51.4% | 51.2% | GPT-5.4 leads |
| AA-GPQA DiamondSource | 92.0% | 91.1% | GPT-5.4 leads |
| AA-HLESource | 41.6% | 37.2% | GPT-5.4 leads |
| AA-Omniscience IndexSource | 5.7% | -11.2% | GPT-5.4 leads |
| AA-Omniscience AccuracySource | 50.0% | 41.5% | GPT-5.4 leads |
| AA-Omniscience Hallucination RateSource | 88.6% | 90.1% | GPT-5.4 leads |
| HealthBench ProfessionalSource | 48.1% | 55.7% | GPT-5.6 Luna leads |
MathGPT-5.6 Luna wins3 benchmarks
MultimodalGPT-5.6 Luna wins11 benchmarks
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Result |
|---|---|---|---|
| MMMU-ProSource | 81.2% | 78.4% | GPT-5.4 leads |
| OfficeQA ProSource | 53.2% | — | Not comparable |
| MMMU-Pro w/ PythonSource | 82.1% | 79.5% | GPT-5.4 leads |
| CharXivSource | 82.8% | — | Not comparable |
| ERQASource | 65.4% | — | Not comparable |
| SimpleVQASource | 61.1% | — | Not comparable |
| ScreenSpot ProSource | 85.4% | — | Not comparable |
| ZeroBenchSource | 41.0% | — | Not comparable |
| MedXpertQA (MM)Source | 77.1% | — | Not comparable |
| AA-MMMU-ProSource | 78.4% | 78.6% | GPT-5.6 Luna leads |
| Design Arena WebsiteSource | 1250 | — | Not comparable |
Inst. Following1 benchmarks
| Benchmark | GPT-5.4 | GPT-5.6 Luna | Result |
|---|---|---|---|
| AA-IFBenchSource | 73.9% | — | Not comparable |
Frequently Asked Questions (6)
Which is better, GPT-5.4 or GPT-5.6 Luna?
GPT-5.4 is ahead on BenchLM's BenchAlign leaderboard, 74.24 to 67.17. The biggest single separator in this matchup is FrontierMath v2 (Tier 4), where the scores are 27.100% and 58.500%.
Which is better for knowledge tasks, GPT-5.4 or GPT-5.6 Luna?
GPT-5.6 Luna has the edge for knowledge tasks in this comparison, averaging 92.3 versus 57.6. Inside this category, AA-Omniscience Index is the benchmark that creates the most daylight between them.
Which is better for coding, GPT-5.4 or GPT-5.6 Luna?
GPT-5.6 Luna has the edge for coding in this comparison, averaging 62.7 versus 57.7. Inside this category, SWE-bench Pro is the benchmark that creates the most daylight between them.
Which is better for math, GPT-5.4 or GPT-5.6 Luna?
GPT-5.6 Luna has the edge for math in this comparison, averaging 73.6 versus 42.5. Inside this category, FrontierMath v2 (Tier 4) is the benchmark that creates the most daylight between them.
Which is better for agentic tasks, GPT-5.4 or GPT-5.6 Luna?
GPT-5.6 Luna has the edge for agentic tasks in this comparison, averaging 84.1 versus 77.2. Inside this category, GDPval-AA is the benchmark that creates the most daylight between them.
Which is better for multimodal and grounded tasks, GPT-5.4 or GPT-5.6 Luna?
GPT-5.6 Luna has the edge for multimodal and grounded tasks in this comparison, averaging 78.4 versus 73.2. Inside this category, MMMU-Pro is the benchmark that creates the most daylight between them.
Related Comparisons
Explore More
Choose a model with this week’s evidence
Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.
One email each week. Unsubscribe anytime.