Skip to content

OpenAI

GPT-5.6 Sol

Introduced OpenAI's three tiers: Sol (frontier), Terra (balanced) and Luna (efficient). Trusted partners got a limited preview first, at the US government's request, before general release on 9 July 2026.

Announced
Released
Access
Public

Benchmark scores

TestScoreBest on the trackerHow it was measured
Epoch Capabilities Indexmany tests combined into one score163 pointsGPT-6.2 Astra 174 pointsIndependentlisted as "GPT-5.6 Sol", CI 160.72 to 165.59Run 9 Nov 2026
FrontierSWE V2very long engineering jobs32%Claude Fable 5.2 70%Independentproximus harness, max effort, mean@5 over 34 tasks, 20-hour budget (best@5 44.2)
Terminal-Bench 4.1long command-line tasks38%Claude Opus 5.5 66%IndependentCodex harness, max effort, 330 trials, ±3.83 (95% CI)
MirrorCoderebuilding whole programs20%Claude Opus 5.5 79%Independenthigh reasoning effort, MirrorCode (ML, +Private, 2L): 15 programs x 2 languages, 3 runs each, 10B-token / 7-day budget, ±9.2 SERun 10 Aug 2026
APEX-Agentsprofessional office tasks51%Gemini 4 Carbon 84%Independentpass@1, default effort; listed as "GPT-5.6 Sol"
ARC-AGI-3puzzle games, standard harnessTypical first-time human player: 100%7.8%Claude Fable 5.2 81%IndependentARC Prize config "GPT-5.6 Sol (Max)" (openai-gpt-5-6-sol-max), standard harness
FrontierMath T1–3hard maths problems89%Claude Fable 5.2 95%Independentmax reasoning effort (gpt-5.6-sol_max)Run 9 Jul 2026
GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated94%Gemini 4 Argon 96%Independentmax reasoning effort (gpt-5.6-sol_max)Run 9 Jul 2026
FrontierMath T4research-level maths problemsSaturated83%GPT-6.1 Sol 100%Independentmax reasoning effort (gpt-5.6-sol_max)Run 9 Jul 2026
ARC-AGI-2visual puzzles, model aloneAverage person: 60%Saturated93%GPT-6 Astra 95%IndependentARC Prize config "GPT-5.6 Sol (Max)" (openai-gpt-5-6-sol-max)
OTIS Mock AIMEcompetition mathsSaturated100%Top scoreIndependentmax reasoning effort (gpt-5.6-sol_max)Run 9 Jul 2026
METR time horizon (TH1.1 + TH2)task length an AI agent finishes half the time12 hGPT-6 Astra 1.7 daysIndependentPooled TH1.1 + TH2 suites; 95% CI 393.4–1239.2 min

Compare every model on the same tests →

Also on these charts

Sources