Atlas

Models

OpenAI

GPT-6.2 Sol

167.8
Capability · 95% interval 165.9–170.0
Rank 7 of 899 (interval 5–10)
Relative cost 0.66× · $0.053 per short-answer task · $0.55 per agentic task

GPT-6.2 Sol is OpenAI's November 17, 2026 efficient reasoning model for everyday work, coding and agents at scale. It accepts text and image inputs and produces text, with low, medium, high, xhigh and max reasoning effort; the API rejects effort none. Tool calling uses the Responses API. Its documented total context window is 1,050,000 tokens, with at most 922,000 input tokens and 128,000 output tokens. It keeps GPT-6.1 Sol's list price of $2 input, $0.10 cached input and $10 output per million tokens.

Scored on 198 benchmarks · 387 total scores · 343 independently reported

Score providers

Organizations reporting headline scores for 198 benchmarks

GPT-6.2 Sol benchmark scores

In fitin the capability model (27 of 198)
Legal · 69 benchmarks · first place on 1
HLAB-Vals-1.1 - Corporate M&A
1 / 40———
HLAB-Vals-1.1 - White Collar Defense Investigations
2 / 40———
Harvey LAB-AA v1.1 (hallucination-gated)
3 / 28———
HLAB-Vals-1.1 - Employment Labor
3 / 40———
HLAB-Vals-1.1 - Immigration
4 / 40———
Finance & Accounting · 35 benchmarks
Excel Modeling Benchmark v1.1 M&A
4 / 44———
Excel Modeling Benchmark v1.1 LBO
6 / 44———
Excel Modeling Benchmark v1.1 DCF
7 / 44———
Excel Modeling Benchmark v1.1 Scratch
8 / 44———
Excel Modeling Benchmark v1.1 Comps
9 / 44———
Code · 16 benchmarks · first place on 3
Terminal-Bench 4.0 — Software
1 / 59———
Terminal-Bench 4.0 — Security
1 / 59———
Code Migration v1.1 (Vals Index v2.2 subset)
1 / 36———
Terminal-Bench 4.0 — Media
2 / 59———
KernelGen1P - OpenAI release percentagenot reported
2 / 8——2026-11-17
Indexes · 14 benchmarks
Artificial Analysis Intelligence Index v5.0.1
3 / 67———
Artificial Analysis Strategy & Ops Index
3 / 26———
Artificial Analysis Engineering Index
3 / 26———
Artificial Analysis Economics Index
3 / 26———
Vals Science Index
4 / 24———
Science · 12 benchmarks
Terminal-Bench Science — Earth Sciences (Vals)
2 / 44———
Terminal-Bench Science — Engineering Sciences (Vals)
2 / 44———
Terminal-Bench-Science 0.1In fit
3 / 78~1 mo~Mar 20272026-11-17
BioMysteryBench — Hard (Vals)
3 / 31———
Terminal-Bench Science 0.1 (Vals)In fit
3 / 44~1 mo~Apr 2027—
General QA · 11 benchmarks
AA-Omniscience Business Accuracy
2 / 26———
AA-Omniscience Health Accuracy
2 / 26———
AA-Omniscience v1.1 Accuracy
3 / 76———
AA-Omniscience Law Accuracy
3 / 26———
ARC-AGI-3 (Semi-Private)In fit
4 / 30~2 mo~Jan 2027—
Agents · 9 benchmarks
LifeVal (GPT-6.2 launch chart)
2 / 4——2026-11-17
Agents' Last Exam v1.1 Score
5 / 19——2026-11-17
AutomationBench-AA Operations, HR, Marketing, Sales and Support
7 / 26———
AutomationBench 1.0.6 (Zapier)In fithigh
8 / 42~2 mo~Sep 20272026-11-17
AutomationBench-AA Finance
10 / 26———
Chat & Writing · 8 benchmarks
AA-PersonalityBench Conviction indexlow
7 / 43———
AA-PersonalityBench Warmth indexx-high
21 / 43———
AA-PersonalityBench Imagination index
21 / 43———
AA-PersonalityBench Transparency index
22 / 43———
AA-PersonalityBench Expressiveness indexmedium
25 / 43———
Cybersecurity · 8 benchmarks · first place on 2
SEC-Bench Pro - release-reported cohortIn fitnot reported
1 / 16~1 moSep 2026*2026-11-17
ExploitBench - release-reported exploit progressnot reported
1 / 18——2026-11-17
DeepsecBench-AA (Artificial Analysis)
2 / 26———
ExploitBench - Internal Port (June-August 2026)
2 / 7——2026-11-17
SecureSWE (OpenAI internal)
2 / 4——2026-11-17
Professional Work · 8 benchmarks · first place on 1
MLCR-AA Accuracy (October 2026 methodology)
1 / 114———
GDP.pdf (Artificial Analysis, Mean Criteria)
2 / 162———
GDPval v2 (OpenAI internal)
3 / 7——2026-11-17
AA-AnalystAgent (pass^5)In fit
7 / 53~2 mo~Dec 2027—
GDP.pdf (Artificial Analysis, All-pass)In fit
7 / 162~2 mo~Jan 2029—
Safety · 5 benchmarks
MentalHealthBench
10 / 29——2026-11-17
AA-Omniscience Business Non-Hallucination
21 / 26———
AA-Omniscience Law Non-Hallucination
21 / 26———
AA-Omniscience Health Non-Hallucination
23 / 26———
AA-Omniscience v1.1 Hallucination Rate
50 / 76———
Health & Medicine · 3 benchmarks
HealthBench Professional (Length-Adjusted)not reported
4 / 60——2026-11-17
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)In fit
7 / 114~4 mo~Mar 2029—
HealthBench ProfessionalIn fitnot reported
10 / 40~3 mo~Aug 20282026-11-17

GPT-6.2 Sol head to head

Related models

Loading Atlas data…