OpenAI · Model release
Data as of September 24, 2026 · How the score is built
GPT-5.6 Luna
Decision readingGPT-5.6 Luna scores 65.6 out of 100 and ranks #25 of 194. This profile shows 30 source-displayable benchmark rows; its strongest eligible category is Coding at #9. API pricing is $0.2 input and $1.2 output per million tokens, with cached input at $0.02.
Released Jul 9, 2026 — see all recent releases
GPT-5.6 Luna will be repriced, updated or retired. Get each notice with its source and date. Follow model changes
Share or export
Decision snapshot
Each value carries a field reference instead of floating alone. Markers compare this model with the current ranked and priced catalog; they are not absolute quality thresholds.
Capability
65.6/100
field median 50.2#25 of 194 ranked models
Public
#25of 194
Verified #17 of 74
Price
$0.20input / $1.20 output
input median $0.97cached $0.020 · blended $0.70
Speed
Not measured
field median 91 tok/sTime to first token not measured
Context
1.05Mtokens
field median 256,000Maximum output length is tracked separately
Strongest published evidence
Coding ranks #9. Particularly well-suited for software development and code generation tasks.
Validate before choosing
30 published rows leave some tracked benchmark slots empty. Independent runtime speed has not been measured.
Source-linked · 30 displayable benchmark rows
Follow model changesCheck the published record for the model and serving route you use.
Check this modelCategory score record
Scores and ranks appear only where published evidence can be displayed. The table keeps the score, weight, cohort, and evidence state together.
| Category | Score | Rank | Percentile | Weight | Benchmarks | Evidence |
|---|---|---|---|---|---|---|
| AgenticRank #28 of 105Percentile 74thWeight 22%9 benchmarksVerified | 55.3 | 9 benchmarks | Verified | |||
| CodingRank #9 of 135Percentile 94thWeight 20%7 benchmarksVerified | 64.5 | 7 benchmarks | Verified | |||
| ReasoningRank #18 of 19Percentile 6thWeight 17%2 benchmarksVerified | 54.7 | 2 benchmarks | Verified | |||
| MultimodalRank #22 of 50Percentile 57thWeight 12%2 benchmarksVerified | 67.1 | 2 benchmarks | Verified | |||
| KnowledgeRank #22 of 158Percentile 87thWeight 12%6 benchmarksVerified | 64.6 | 6 benchmarks | Verified | |||
| MultilingualWeight 7%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| Inst. FollowingWeight 5%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| MathRank Not rankedWeight 5%3 benchmarksVerified | 94.2 | 3 benchmarks | Verified |
30 of 483 tracked benchmark slots have displayable evidence · bars run 0–100
Coverage detailsHow much of this is verified
Coverage is split by category so a strong number never hides a thin evidence base. Verified means the row is tied to a published source; provisional rows remain visible but separate.
- Agentic9/9 verified
- Coding7/7 verified
- Reasoning2/2 verified
- Multimodal2/2 verified
- Knowledge6/6 verified
- MultilingualNot measured
- Inst. FollowingNot measured
- Math3/3 verified
Capability shape
Each axis shows percentile within that category’s eligible cohort. The comparison outline is the median of the six nearest public-score peers; a collapsed vertex means the category is not rank-eligible.
GPT-5.6 Luna category percentile values
- Agentic74th percentile
- Coding94th percentile
- Reasoning6th percentile
- Multimodal57th percentile
- Knowledge87th percentile
- MultilingualNot eligible
- Instruction followingNot eligible
- MathNot eligible
The dashed outline is median of 6 nearest peers.
Eligible category ranks
- Agentic#28/105
- Coding#9/135
- Reasoning#18/19
- Multimodal#22/50
- Knowledge#22/158
- MultilingualNot ranked
- Inst. FollowingNot ranked
- MathNot ranked
Benchmark ledger
Coding opens by default. The marker compares each value with the best source-verified result in the catalog; provisional leaders do not set the reference. Expand the remaining categories for every published row.
Coding7 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| SWE-bench Pro | Score62.7% | Versus best verified row Best verified: Claude Opus 5.5 · 89.9% | Gap27.2 behind | Weight26% ref. weight | Provider exact |
| DeepSWE | Score67.2% | Versus best verified row Best verified: Muse Spark 1.3 · 75.4% | Gap8.2 behind | Weight15% ref. weight | Provider exact |
| CursorBench 3.2 | Score61.1% | Versus best verified row Best verified: Claude Fable 5.1 · 73.4% | Gap12.3 behind | Weight10% ref. weight | Benchmark exact |
| VulcanBench v3 | Score85.5% | Versus best verified row Best verified: Grok 4.5 · 89.9% | Gap4.4 behind | Weight3% ref. weight | Benchmark exact |
| Terminal-Bench 2.1 | Score84.7% | Versus best verified row Best verified: SWE-2 · 92.8% | Gap8.1 behind | WeightScored in Agentic | Provider exact |
| FrontierCode 1.1 Extended | Score55.1% | Versus best verified row Best verified: GPT-6 Astra · 64.5% | Gap9.4 behind | WeightDisplay only | Provider exact |
| SWE-bench (Vals)SWE-bench, Vals AI run | Score93.0% | Versus best verified row Best verified: Claude Opus 5 · 97.0% | Gap4 behind | WeightDisplay only | Verified |
Agentic9 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| OSWorld 2.0 | Score45.6% | Versus best verified row Best verified: GPT-6 Astra · 72.6% | Gap27 behind | Weight10% ref. weight | Provider exact |
| Terminal-Bench 2.1 | Score84.7% | Versus best verified row Best verified: SWE-2 · 92.8% | Gap8.1 behind | Weight8% ref. weight | Provider exact |
| BrowseComp | Score83.3% | Versus best verified row Best verified: Atria Dawn Preview · 92.5% | Gap9.2 behind | Weight8% ref. weight | Provider exact |
| Terminal-Bench 2.1 (Vals)Terminal-Bench 2.1, Vals AI run | Score79.0% | Versus best verified row Best verified: GPT-6 Astra · 87.3% | Gap8.3 behind | Weight3% ref. weight | |
| Terminal-Bench 3.0 | Score14.3% | Versus best verified row Best verified: Claude Opus 5 · 42.7% | Gap28.4 behind | Weight3% ref. weight | Benchmark exact |
| Toolathlon | Score53.4% | Versus best verified row Best verified: Muse Spark 1.1 · 75.6% | Gap22.2 behind | Weight3% ref. weight | Provider exact |
| CyberGym | Score77.9% | Versus best verified row Best verified: MiMo-V2.6-Flash · 95.1% | Gap17.2 behind | WeightDisplay only | Provider exact |
| ExploitGym | Score12.4% | Versus best verified row Best verified: GPT-6 Astra · 42.4% | Gap30 behind | WeightDisplay only | Provider exact |
| ApprenticeBenchApprenticeBench: end-to-end computer use, continual learning, and long-horizon agency on a real accounts-payable job | Score7% | Versus best verified row Best verified: Claude Fable 5.1 · 72% | Gap65 behind | WeightDisplay only | Benchmark exact |
Reasoning2 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| ARC-AGI-2Abstraction and Reasoning Corpus for AGI v2 | Score59.5% | Versus best verified row Best verified: GPT-6 Astra · 95% | Gap35.5 behind | WeightWeighted 25% | Benchmark exact |
| ARC-AGI-3Abstraction and Reasoning Corpus for AGI v3 | Score0.2% | Versus best verified row Best verified: GPT-6 Astra · 62.7% | Gap62.5 behind | WeightWeighted 15% | Benchmark exact |
Multimodal2 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| MMMU-ProMassive Multi-discipline Multimodal Understanding Pro | Score78.4% | Versus best verified row Best verified: Gemini 3.5 Flash · 83.6% | Gap5.2 behind | WeightWeighted 40% | Provider exact |
| MMMU-Pro w/ PythonMMMU-Pro with Python | Score79.5% | Versus best verified row Best verified: GPT-5.6 Sol · 84.6% | Gap5.1 behind | WeightDisplay only | Provider exact |
Knowledge6 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| MMLU-Pro (Vals)MMLU-Pro, Vals AI run | Score86.0% | Versus best verified row Best verified: Claude Fable 5.1 · 92.4% | Gap6.4 behind | Weight6% ref. weight | Verified |
| GPQAGraduate-Level Google-Proof Q&A | Score92.3% | Versus best verified row Best verified: GPT-6 Astra · 96% | Gap3.7 behind | Weight3% ref. weight | Provider exact |
| GPQA Diamond (Vals)GPQA Diamond, Vals AI run | Score91.7% | Versus best verified row Best verified: Gemini 3.1 Pro · 95.5% | Gap3.8 behind | Weight2% ref. weight | |
| GPQA-DGPQA Diamond | Score92.3% | Versus best verified row Best verified: GPT-6 Astra · 96.0% | Gap3.7 behind | WeightDisplay only | Provider exact |
| HealthBench Professional | Score55.7% | Versus best verified row Best verified: Claude Opus 5.5 · 65.6% | Gap9.9 behind | WeightDisplay only | Provider exact |
| HealthBench Hard | Score32.0% | Versus best verified row Best verified: Muse Spark · 42.8% | Gap10.8 behind | WeightDisplay only | Provider exact |
Math3 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| FrontierMath v2 (Tiers 1-3)FrontierMath v2 Tiers 1-3 | Score78.600% | Versus best verified row Best verified: GPT-5.6 Sol · 89.000% | Gap10.4 behind | WeightWeighted 30% | Provider exact |
| FrontierMath v2 (Tier 4)FrontierMath v2 Tier 4 | Score58.500% | Versus best verified row Best verified: GPT-6 Astra · 97.600% | Gap39.1 behind | WeightWeighted 10% | Provider exact |
| FrontierMath (legacy)FrontierMath legacy aggregate | Score78.6% | Versus best verified row Best verified: GPT-5.6 Sol · 89% | Gap10.4 behind | WeightDisplay only | Provider exact |
External signals1 row
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| ExploitBenchExploitBench v8-bench | Score33% | Versus best verified row Best verified: GPT-6 Astra · 100% | Gap66.8 behind | WeightDisplay only | Provider exact |
What it costs to get this score
Published API price against the public score. The x-axis uses a log scale; the dashed path marks models that are not beaten by a cheaper, higher-scoring option. Price uses average of published input and output rates.
GPT-5.6 Luna · 65.6 score · $0.70 blended per million tokens
The chart opens on the current model. Scroll horizontally to inspect the full price axis.
Horizontal: blended price per million tokens, log scale · Vertical: public score
Published price history
This chart appears only when at least two dated first-party price records exist for the exact model. It shows rate changes, not an inferred cost trend.
2 dated first-party price records. The newest record lists $1.2 per million tokens.
Lineage
The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.
GPT-5.6 Luna release history
Radar confirmed these at the source. Use GPT-5.6 Luna in your work? Explore Radar to follow supported changes and choose your alerts.
Radar
Spec sheet
Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.
- API model ID
- gpt-5.6-lunaOpenAI model catalog
- Context window
- 1.05MOpenAI model catalog
- Maximum output
- Not sourced yet
- Knowledge cutoff
- Not sourced yet
- Input modalities
- text, imageOpenAI model catalog
- Output modalities
- textOpenAI model catalog
- Parameters
- Not disclosed by the provider
- Availability
- OpenAI Responses APIOpenAI model catalog
- Cloud regions
- Not tracked yet
- Lifecycle
- activeOpenAI model catalog
- API capabilities
- Tool calling, structured outputs, and batch support are not tracked yet
- Prompt caching
- Published at $0.020 per million cached input tokensOpenAI pricing
- Self-host
- Weights are not published
- Rate limits
- Not tracked yet
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
GPT-5.6 Luna ranks #25 of 194 on the public leaderboard with a score of 65.6/100. Its source-verified position is #17 of 74.
GPT-5.6 Luna is a proprietary model with a 1.05M context window. It uses an explicit reasoning mode, which can improve complex problem solving while adding latency and token use.
GPT-5.6 Luna sits in the GPT-5.6 family with GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Cyber. 30 of 483 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Its strongest eligible category is Coding at #9, while its lowest eligible position is Agentic at #28. particularly well-suited for software development and code generation tasks.
Last updated September 24, 2026. Runtime fields remain blank until a sourced snapshot exists.
Questions
How does GPT-5.6 Luna perform overall in AI benchmarks?
GPT-5.6 Luna ranks #25 out of 194 models on the public BenchAlign leaderboard, with a score of 65.6/100. Its evidence status is Supported, and this profile shows 30 source-displayable benchmark rows. The label describes evidence depth, not a provider quality claim; inspect category rows before choosing a workload.
Is GPT-5.6 Luna good for knowledge and understanding?
GPT-5.6 Luna ranks #22 out of 158 eligible models for knowledge and understanding, with a public category score of 64.6/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is GPT-5.6 Luna good for coding and programming?
GPT-5.6 Luna ranks #9 out of 135 eligible models for coding and programming, with a public category score of 64.5/100. That places it in the current top ten for this category. Check the underlying rows before treating the aggregate as a workload guarantee.
Is GPT-5.6 Luna good for mathematics?
GPT-5.6 Luna has source-displayable benchmark coverage for mathematics, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is GPT-5.6 Luna good for reasoning and logic?
GPT-5.6 Luna ranks #18 out of 19 eligible models for reasoning and logic, with a public category score of 54.7/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is GPT-5.6 Luna good for agentic tool use and computer tasks?
GPT-5.6 Luna ranks #28 out of 105 eligible models for agentic tool use and computer tasks, with a public category score of 55.3/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is GPT-5.6 Luna good for multimodal and grounded tasks?
GPT-5.6 Luna ranks #22 out of 50 eligible models for multimodal and grounded tasks, with a public category score of 67.1/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Which sibling models are related to GPT-5.6 Luna?
GPT-5.6 Luna belongs to the GPT-5.6 family. Related tracked variants include GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Cyber. A sibling link indicates shared lineage or a documented configuration relationship; it does not mean the variants have identical pricing, context limits, benchmark evidence, or deployment behavior. Compare before switching.
Does GPT-5.6 Luna have full benchmark coverage on BenchLM?
No. GPT-5.6 Luna currently has 51 source-displayable rows across 483 tracked benchmark slots. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.
What is the context window size of GPT-5.6 Luna?
GPT-5.6 Luna has a documented context window of 1.05M. That figure is the maximum combined prompt and retained-conversation space reported for this exact model; it is not the maximum output length. The profile keeps output limits separate because providers often publish those limits independently.