OpenAIReasoning modelNewReleased 22 Sept 2026

GPT-6 Luna

Reasoning effort

GPT-6 Luna is an OpenAI proprietary reasoning model, released 22 Sept 2026. Its best configuration (max reasoning effort) ranks #92 of 432 on the BenchLeader Index at 61.2 ±5.6, in the upper half. It scores highest in reasoning (76) and lowest in multimodal (59). At $0.200 per million tokens blended it is cheaper than most ranked models. Output speed of 157 tokens per second puts it in the fastest quarter, with a first answer in 142.6 s. It has been measured at 6 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 22 Sept 2026.

Blended price: $0.200/M; $0.100 in · $0.500 out
Output speed: 157 tok/s; measured by Artificial Analysis
First answer: 143 s; first token 1.65 s
Context: 1.1M
Full answer: 146 s; median, reasoning included
Cost per run: $0.068; one full Intelligence Index run
Released: 22 Sept 2026

How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.

Overall index61
Reasoning76
Knowledge63
Multimodal59
Long context68
Composite74

Versions

OpenAI has shipped 2 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.

Model	Released	Index	Rank
NewGPT-6 Lunamaxthis page	22 Sept 2026	61.2	#92
GPT-5.6 Lunaxhigh	9 Jul 2026	60.6	#101

Reasoning-effort configurations

The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
no reasoning	47.3	#428	141 tok/s	0.71 s	$0.0002	Composite 51 · Knowledge 53 · Long context 45 · Multimodal 36 · Reasoning 44
low	53.2	#268	175 tok/s	1.59 s	$0.0002	Composite 54 · Knowledge 59 · Long context 63 · Multimodal 54 · Reasoning 46
medium	57.1	#172	143 tok/s	5.28 s	$0.0002	Composite 65 · Knowledge 61 · Long context 65 · Multimodal 57 · Reasoning 60
high	58.8	#136	149 tok/s	7.13 s	$0.0002	Composite 68 · Knowledge 60 · Long context 65 · Multimodal 58 · Reasoning 69
xhigh	59.9	#115	153 tok/s	16 s	$0.0002	Composite 70 · Knowledge 62 · Long context 66 · Multimodal 59 · Reasoning 73
maxbest	61.2	#92	157 tok/s	143 s	$0.0002	Composite 74 · Knowledge 63 · Long context 68 · Multimodal 59 · Reasoning 76

Benchmark results

One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Reasoning

Benchmark	no reasoning	low	medium	high	xhigh	max	Source
Humanity's Last Exam (AA)not in index	8.6%#312	20.3%#183	28.3%#137	33.0%#113	34.3%#101	38.5%#78	Artificial Analysis
CritPt	1.1%#190	2.6%#159	10.6%#91	15.4%#70	17.4%#61	19.4%#50	Artificial Analysis

Coding

Benchmark	no reasoning	low	medium	high	xhigh	max	Source
SciCode (AA)not in index	43.1%#125	46.9%#105	50.9%#79	50.4%#84	51.7%#70	54.6%#49	Artificial Analysis
Terminal-Bench 4.0 (AA)not in index	1.5%#92	0.0%#120	2.5%#84	4.5%#77	8.1%#66	12.6%#53	Artificial Analysis

Agents & tools

Benchmark	no reasoning	low	medium	high	xhigh	max	Source
GDPval (AA)not in index	26.7%#115	24.6%#124	35.9%#90	39.5%#78	39.8%#75	43.4%#60	Artificial Analysis
AutomationBenchnot in index	–	–	–	–	–	53.2%#35	Artificial Analysis
GDP.pdfnot in index	–	–	–	–	–	20.4%#25	Artificial Analysis
AA-Briefcasenot in index	–	–	–	–	–	1299#37	Artificial Analysis

Knowledge

Benchmark	no reasoning	low	medium	high	xhigh	max	Source
AA-Omniscience	-21.9#215	-9.2#156	-5.0#136	-5.5#139	-1.8#120	0.7#106	Artificial Analysis
AA-Omniscience: accuracynot in index	31.9%#142	40.8%#99	43.1%#85	42.8%#88	44.2%#80	43.8%#81	Artificial Analysis
AA-Omniscience: non-hallucinationnot in index	21.1%#243	15.7%#301	15.3%#305	15.6%#302	17.6%#283	23.3%#227	Artificial Analysis

Multimodal

Benchmark	no reasoning	low	medium	high	xhigh	max	Source
MMMU-Pro	53.2%#219	70.3%#131	73.0%#115	74.4%#98	75.3%#89	75.5%#81	Artificial Analysis

Long context

Benchmark	no reasoning	low	medium	high	xhigh	max	Source
AA-LCR	39.7%#322	74.0%#142	78.3%#95	79.3%#81	80.0%#66	83.3%#18	Artificial Analysis
MLCRnot in index	–	–	–	–	–	16.1%#24	Artificial Analysis

Composite

Benchmark	no reasoning	low	medium	high	xhigh	max	Source
AA Intelligence Index	18.3#223	20.9#189	29.5#101	32.1#87	33.9#75	37.3#60	Artificial Analysis

Where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

Provider	Speed	First token	Input $/M	Output $/M	Context	Quantisation
Azure	127 tok/s	2.71 s	$0.110	$0.550	1.1M	–
OpenAI	119 tok/s	2.52 s	$0.200	$1.00	1.1M	–
Azure	117 tok/s	4.10 s	$0.100	$0.500	1.1M	–
Azure	105 tok/s	1.22 s	$0.110	$0.550	1.1M	–
OpenAI	99 tok/s	2.53 s	$0.050	$0.250	1.1M	–
OpenAI	57 tok/s	1.65 s	$0.100	$0.500	1.1M	–
Amazon Bedrock	44 tok/s	0.86 s	$0.110	$0.550	1.1M	–

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.010 per 1M. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0002	$0.0002	2.4 min
Summarise a 30-page report	12,000 / 600	$0.0015	$0.0007	2.4 min
Code edit	6,000 / 1,500	$0.0014	$0.0009	2.5 min
Agentic coding session	60,000 / 4,000	$0.0080	$0.0040	2.8 min
Structured extraction	2,000 / 200	$0.0003	$0.0002	2.4 min