GPT-6 Luna
OpenAI · released
The smallest and cheapest GPT-6 model, meant for narrow, well-scoped, high-volume jobs such as extraction, classification, triage and small edits — while keeping the 1M-token context window and the full tool set.
Best for: High-volume automation · Classification and extraction · Sub-agents
- Intelligence Index
- 38.1
- #21 of 38 · at Max effort
- Price per 1M tokens
- $0.10 in $0.50 out
- Blended $0.20 at 3:1
- Context window
- 1.05M
- Up to 128K output
- Output speed
- 129 tokens/s
- $0.07 measured cost per task
Capability profile
One 0–100 score per area, placing this model between the weakest and strongest tracked model across that area's benchmarks.
- Overall58 · #21 of 38
Based on 3 of 3 benchmarks
- Coding50 · #20 of 39
Based on 5 of 5 benchmarks
- Agents & knowledge work55 · #24 of 40
Based on 6 of 6 benchmarks
- Reasoning & science58 · #19 of 39
Based on 2 of 2 benchmarks
- Knowledge & reliability66 · #18 of 39
Based on 2 of 2 benchmarks
- Vision74 · #13 of 31
Based on 1 of 2 benchmarks
Reasoning effort trade-off
The same model at each effort setting. Lower settings give up a few points for a fraction of the cost and wait.
| Setting | Intelligence Index | Cost / task | First answer |
|---|---|---|---|
| Max | 38.1 | $0.07 | 1.9 min |
| xHighSweet spot | 34.6 | $0.04 | 36s |
| High | 32.9 | $0.03 | 21s |
| Medium | 29.9 | $0.02 | — |
| Low | 21.5 | $0 | 3.1s |
| No reasoning | 18.5 | $0.01 | 0.8s |
Measured by Artificial Analysis. First answer is the median wait until the first answer token, including thinking time. “Sweet spot” is the cheapest setting that keeps at least 90% of the top score for no more than 60% of its cost.
Benchmark results
All third-party results on record. Rank is among the tracked models that have been tested on that benchmark.
Overall
#21 of 38· Max effort· Artificial Analysis
- Vals Index51.2%
#16 of 26· Max effort· $0.43 per test· Vals AI
- Arena · Text1443 ±7
#23 of 33· Max effort· 10,014 votes· Arena
Coding
- Terminal-Bench 4.012.6%
#22 of 38· Max effort· Artificial Analysis
- SciCode54.6%
#16 of 38· Max effort· Artificial Analysis
- Arena · WebDev1582 ±7
#20 of 34· Max effort· 10,729 votes· Arena
- Vibe Code Bench81.7%
#14 of 26· Max effort· Vals AI
- Code Migration44.4%
#12 of 26· Max effort· Vals AI
Agents & knowledge work
- GDPval-AA46.9
#21 of 39· Max effort· Artificial Analysis
- Arena · Agent+1.4 ±1.1
#16 of 29· Max effort· Arena
- Finance Agent v249.9%
#22 of 26· Max effort· Vals AI
#8 of 26· Max effort· Vals AI
- Legal Research Bench30.3%
#22 of 26· Max effort· Vals AI
- Tax Agent Bench58.9%
#21 of 26· Max effort· Vals AI
Reasoning & science
- Humanity's Last Exam38.5%
#22 of 39· Max effort· Artificial Analysis
- CritPt19.4%
#12 of 39· Max effort· Artificial Analysis
- ARC-AGI-2saturating59.3%
#13 of 16· Max effort· $0.06 per task· ARC Prize
- ARC-AGI-30.2%
#6 of 6· Medium effort· ARC Prize
- GPQA DiamondsaturatingNot tested
Knowledge & reliability
#20 of 39· Max effort· Artificial Analysis
- AA-LCR83.3%
#7 of 39· Max effort· Artificial Analysis
Vision
- MMMU-Pro79.7%
#11 of 27· Max effort· Artificial Analysis
- Arena · VisionNot tested
Reported by OpenAI
Numbers the lab published at launch, run on its own harness and settings. Useful context, but not independently verified and never used in the rankings on this site.
- DeepSWE v1.1 · Max effort66.6%
- FrontierCode 1.1 Main · Max effort42.4%
- OSWorld 2.0 · offline set, partial score52.7%
Specifications
- Lab
- OpenAI
- Released
- Sep 22, 2026
- Availability
- Generally available
- Context window
- 1,050,000 tokens
- Max output
- 128,000 tokens
- Accepts
- Text, Images, PDFs
- Knowledge cutoff
- May 2026
- Weights
- Closed
- Reasoning control
- Effort none to max · pro mode
- API model ID
gpt-6-luna
Pricing
- Input / 1M tokens
- $0.10
- Cached input / 1M
- $0.01
- Output / 1M tokens
- $0.50
- Blended (3:1)
- $0.20
- Measured cost / task
- $0.07
Prompts over 272K input tokens are billed at $0.20 in / $0.75 out for the whole request. Batch and Flex are half price.
Official sources
Closest alternatives
The models nearest to this one on overall score.
23 benchmarks tracked. See methodology and sources.