GPT-6 Luna

Data updated 31 Oct 2026

21 published benchmark measures · 2 benchmark families contribute across 2 task areas. 2 capability estimates available in the full profile. See all results ↓ · Compare published benchmarks →

Model evidence summary

This profile combines published settings. It is not a runnable configuration or a leaderboard rank. Compare measured configurations →

Performance profile

Capabilities

Adjusted comparison score · 0–100

Filled points are supported; hollow points are preliminary. Lines stop at unknown capabilities. Exact values and sources follow below.

AgenticNo comparable evidenceUnknown

0 core families; 0 direct opponents across 0 labs. No admitted core comparisons

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: No supported estimate; 13/13 scenarios unsupported. Without one publisher: No supported estimate; 15/15 unsupported. Smoothing check: No supported estimate; 3/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: family and publisher holdouts improved on neutral predictions within evaluated coverage; source sensitivity still requires inspection. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

Hard reasoning1 families · 1 with independent evidence · Preliminary32.7

1 core families; 59 direct opponents across 10 labs. Needs broader benchmark and opponent support

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: No supported estimate; 6/6 scenarios unsupported. Without one publisher: No supported estimate; 16/16 unsupported. Smoothing check: No supported estimate; 3/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: family and publisher holdouts improved on neutral predictions within evaluated coverage. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

ARC-AGI: 49.2 observed win share
arcprize.org · Source 1

Coding1 families · 1 with independent evidence · Preliminary43.7

1 core families; 21 direct opponents across 7 labs. Needs broader benchmark and opponent support

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: No supported estimate; 6/6 scenarios unsupported. Without one publisher: No supported estimate; 19/19 unsupported. Smoothing check: No supported estimate; 3/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: family and publisher holdouts improved on neutral predictions within evaluated coverage. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

Terminal-Bench: 19.0 observed win share
tbench.ai · Source 1

Human prefNo comparable evidenceUnknown

0 core families; 0 direct opponents across 0 labs. No admitted core comparisons

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: No supported estimate; 1/1 scenarios unsupported. Without one publisher: No supported estimate; 1/1 unsupported. Smoothing check: No supported estimate; 3/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: cross-family and cross-publisher predictions could not be evaluated from the available comparison network. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

KnowledgeNo comparable evidenceUnknown

0 core families; 0 direct opponents across 0 labs. No admitted core comparisons

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: No supported estimate; 4/4 scenarios unsupported. Without one publisher: No supported estimate; 5/5 unsupported. Smoothing check: No supported estimate; 3/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: almost no publisher-held-out comparisons remained estimable. Cross-source validity is unresolved. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

MultimodalNo comparable evidenceUnknown

0 core families; 0 direct opponents across 0 labs. No admitted core comparisons

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: No supported estimate; 8/8 scenarios unsupported. Without one publisher: No supported estimate; 8/8 unsupported. Smoothing check: No supported estimate; 3/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: limited publisher-held-out coverage. Cross-source validity is unresolved. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

Long contextNo comparable evidenceUnknown

0 core families; 0 direct opponents across 0 labs. No admitted core comparisons

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: No supported estimate; 5/5 scenarios unsupported. Without one publisher: No supported estimate; 5/5 unsupported. Smoothing check: No supported estimate; 3/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: family and publisher holdouts underperformed neutral predictions. Treat this as an exploratory estimate. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

Scores estimate outcomes against a shared reference panel; they are not accuracy percentages. Sparse or disconnected evidence cannot qualify an overall profile. Open a capability to inspect its evidence.

Reported effort · Mixed settings

Settings reported in this model's published benchmark results, including results outside the aggregate. Effort names are provider-specific. These are not API defaults or equal compute budgets.

High: 5 observations
Low: 5 observations
Max: 8 observations
Medium: 5 observations
XHigh: 6 observations
Not specified: 14 observations

Mixed settings means multiple settings occur in the evidence. Best across efforts means the source selected its best reported result across settings; it does not mean Max. Unspecified settings stay unknown. This model-summary chart combines reported settings. The leaderboard keeps identified configurations separate and excludes unknown effort.

Inspect each result and its source ↓ · Download effort evidence

Score contributions and missing evidence

2 contributing families across 2 capabilities. Fixed reference panels do not change when the catalog expands.

Capability is fitted jointly across families. Capability estimates below describe different task areas; their weighted sum is not the Capability score.

Results without reviewed compatibility or a reference match remain in the raw evidence below. Coverage counts only contributing results.

Model information & shareable badge

Lab: OpenAI
Catalog status: active
Availability: Public provider catalog; account and region restrictions may apply
Family: GPT-6
Released: 2026-09-22
Context: 1,050,000 tokens
API list price: $0.10 input / $0.50 output per million tokens
Standard rate for requests up to 272,000 input tokens. Longer requests are priced at 2× input and 1.5× output for the whole request.
License: proprietary
Model card: https://developers.openai.com/api/docs/models/gpt-6-luna
Default Capability family coverage: /badge/gpt-6-luna.svg

Benchmark scores & sources

Original results, evaluation harnesses, and evidence behind this model.

Benchmark	Bucket	Score	Harness	Evidence	Source-recorded date
Agents' Last Exam	Supporting evidence	25%	Agents' Last Exam · Codex Effort: Not specified	official board Supporting evidence outside the reviewed capability core	2026-10-20
Agents' Last Exam · V1	Supporting evidence	36.32% Reported settings & source Reported reasoning effort low. Agents' Last Exam V1. Long-horizon professional tasks across 55 sub-industries. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.3632, stored as 36.32 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: Agents' Last Exam / GPT-6 Luna / low · reviewed 2026-09-22	Published configuration Effort: Low	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
Agents' Last Exam · V1	Supporting evidence	46.84% Reported settings & source Reported reasoning effort medium. Agents' Last Exam V1. Long-horizon professional tasks across 55 sub-industries. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.4684, stored as 46.84 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: Agents' Last Exam / GPT-6 Luna / medium · reviewed 2026-09-22	Published configuration Effort: Medium	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
Agents' Last Exam · V1	Supporting evidence	43.6% Reported settings & source Reported reasoning effort high. Agents' Last Exam V1. Long-horizon professional tasks across 55 sub-industries. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.436, stored as 43.6 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: Agents' Last Exam / GPT-6 Luna / high · reviewed 2026-09-22	Published configuration Effort: High	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
Agents' Last Exam · V1	Supporting evidence	47.89% Reported settings & source Reported reasoning effort xhigh. Agents' Last Exam V1. Long-horizon professional tasks across 55 sub-industries. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.4789, stored as 47.89 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: Agents' Last Exam / GPT-6 Luna / xhigh · reviewed 2026-09-22	Published configuration Effort: XHigh	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
Agents' Last Exam · V1	Supporting evidence	50.89% Reported settings & source Reported reasoning effort max. Agents' Last Exam V1. Long-horizon professional tasks across 55 sub-industries. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.5089, stored as 50.89 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: Agents' Last Exam / GPT-6 Luna / max · reviewed 2026-09-22	Published configuration Effort: Max	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
ARC-AGI-2	Hard reasoning	59.3%	ARC Prize verified Effort: Not specified contributes to capability	official board	2026-09-22
Artificial Analysis Cyber Index	Supporting evidence	52.6853594148008	Artificial Analysis Stirrup · Cyber Index v1.0 · max Effort: Max Cyber Index source settings Artificial Analysis Stirrup, v1.0. Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall. Official Cyber Index board	official board Supporting evidence outside the reviewed capability core	2026-10-31
AutomationBench · 1.0.6	Agentic	1.2% Reported settings & source Reported reasoning effort low. AutomationBench 1.0.6. End-to-end workflows across 47 tools in sales, marketing, operations, support, finance, and HR. The page says the Claude Fable 5.1 cost point omits Opus 5 fallbacks. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.012, stored as 1.2 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: AutomationBench / GPT-6 Luna / low · reviewed 2026-09-22	Published configuration Effort: Low	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
AutomationBench · 1.0.6	Agentic	9.4% Reported settings & source Reported reasoning effort medium. AutomationBench 1.0.6. End-to-end workflows across 47 tools in sales, marketing, operations, support, finance, and HR. The page says the Claude Fable 5.1 cost point omits Opus 5 fallbacks. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.094, stored as 9.4 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: AutomationBench / GPT-6 Luna / medium · reviewed 2026-09-22	Published configuration Effort: Medium	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
AutomationBench · 1.0.6	Agentic	14.5% Reported settings & source Reported reasoning effort high. AutomationBench 1.0.6. End-to-end workflows across 47 tools in sales, marketing, operations, support, finance, and HR. The page says the Claude Fable 5.1 cost point omits Opus 5 fallbacks. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.145, stored as 14.5 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: AutomationBench / GPT-6 Luna / high · reviewed 2026-09-22	Published configuration Effort: High	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
AutomationBench · 1.0.6	Agentic	12.6% Reported settings & source Reported reasoning effort xhigh. AutomationBench 1.0.6. End-to-end workflows across 47 tools in sales, marketing, operations, support, finance, and HR. The page says the Claude Fable 5.1 cost point omits Opus 5 fallbacks. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.126, stored as 12.6 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: AutomationBench / GPT-6 Luna / xhigh · reviewed 2026-09-22	Published configuration Effort: XHigh	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
AutomationBench · 1.0.6	Agentic	20.7% Reported settings & source Reported reasoning effort max. AutomationBench 1.0.6. End-to-end workflows across 47 tools in sales, marketing, operations, support, finance, and HR. The page says the Claude Fable 5.1 cost point omits Opus 5 fallbacks. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.207, stored as 20.7 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: AutomationBench / GPT-6 Luna / max · reviewed 2026-09-22	Published configuration Effort: Max	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
AutomationBench-AA	Supporting evidence	53.16814034483175%	Artificial Analysis AutomationBench-AA Effort: Not specified	official board Supporting evidence outside the reviewed capability core	2026-10-30
Bug Hunt Bench	Supporting evidence	17.43%	Bug Hunt Bench · max Effort: Not specified	official board Supporting evidence outside the reviewed capability core	2026-09-23
Bug Hunt Bench	Supporting evidence	13.33%	Bug Hunt Bench · xhigh Effort: Not specified	official board Supporting evidence outside the reviewed capability core	2026-09-23
Computer-use safety stress test · Sol 6.1 launch / safety stress test	Supporting evidence	13.7% Reported settings & source Reported reasoning effort xhigh. Unintended outcomes in deliberately adversarial computer- and browser-use workplace tasks; the updated harder safety subset, not OSWorld task success. OpenAI research environment or API; same named metric and evaluation chart. No fallback is reported for this OpenAI configuration. Provider-published lab self-report. Safety stress outcomes are retained for inspection and excluded from capability scoring. Competitor results are republished from public reports; the page does not establish independent reproduction by OpenAI. Comparison limit: Safety failure behavior is not a capability or intelligence score. Introducing GPT-6.1 Sol · Chart: Computer-use safety stress test (lower is better) / GPT-6 Luna / xhigh · reviewed 2026-09-29	Published configuration Effort: XHigh	lab self-report Safety failure behavior is not a capability or intelligence score.	Reviewed 2026-09-29
DAYJOB Finance	Supporting evidence	6.8%	DAYJOB Finance · max Effort: Not specified	official board Supporting evidence outside the reviewed capability core	2026-10-28
DAYJOB Healthcare	Supporting evidence	1.2%	DAYJOB Healthcare · max Effort: Not specified	official board Supporting evidence outside the reviewed capability core	2026-10-28
DeepSWE v1.1 · v1.1	Coding	2.43% Reported settings & source Reported reasoning effort low. DeepSWE v1.1. Original long-horizon software-engineering tasks. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.0243, stored as 2.43 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: DeepSWE / GPT-6 Luna / low · reviewed 2026-09-22	Published configuration Effort: Low	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
DeepSWE v1.1 · v1.1	Coding	44.47% Reported settings & source Reported reasoning effort medium. DeepSWE v1.1. Original long-horizon software-engineering tasks. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.4447, stored as 44.47 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: DeepSWE / GPT-6 Luna / medium · reviewed 2026-09-22	Published configuration Effort: Medium	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
DeepSWE v1.1 · v1.1	Coding	59.29% Reported settings & source Reported reasoning effort high. DeepSWE v1.1. Original long-horizon software-engineering tasks. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.5929, stored as 59.29 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: DeepSWE / GPT-6 Luna / high · reviewed 2026-09-22	Published configuration Effort: High	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
DeepSWE v1.1 · v1.1	Coding	61.28% Reported settings & source Reported reasoning effort xhigh. DeepSWE v1.1. Original long-horizon software-engineering tasks. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.6128, stored as 61.28 percent. Not an independent board. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: DeepSWE / GPT-6 Luna / xhigh · reviewed 2026-09-22	Published configuration Effort: XHigh	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
DeepSWE v1.1 · v1.1	Coding	66.59% Reported settings & source Reported reasoning effort max. DeepSWE v1.1. Original long-horizon software-engineering tasks. OpenAI-published chart on the Introducing GPT-6 Sol and Luna page. Provider-published lab self-report. Vega-Lite chart data value 0.6659, stored as 66.59 percent. Not an independent board. Article prose rounds this max point to 66.6%. Comparison limit: Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison. Introducing GPT-6 Sol and Luna · Chart: DeepSWE / GPT-6 Luna / max · reviewed 2026-09-22	Published configuration Effort: Max	lab self-report Provider-published lab self-report. Shown as a labeled claim and not admitted as a matched board comparison.	Reviewed 2026-09-22
Failure to disclose a broken search tool · Sol 6.1 launch / safety stress test	Supporting evidence	28.67% Reported settings & source Reported reasoning effort max. Adversarial test of whether agents disclose a broken search tool. OpenAI research environment or API; same named metric and evaluation chart. No fallback is reported for this OpenAI configuration. Provider-published lab self-report. Safety stress outcomes are retained for inspection and excluded from capability scoring. Competitor results are republished from public reports; the page does not establish independent reproduction by OpenAI. Comparison limit: Safety failure behavior is not a capability or intelligence score. Introducing GPT-6.1 Sol · Chart: Failure to disclose a broken search tool (lower is better) / GPT-6 Luna / max · reviewed 2026-09-29	Published configuration Effort: Max	lab self-report Safety failure behavior is not a capability or intelligence score.	Reviewed 2026-09-29