Frontier Log

OpenAI

GPT-6.2 Sol

Released · openai.com

Tier Astra › Sol › Terra › Luna

Price

Per 1M tokensInputCachedOutput
List priceCache writes $2.50 (30 minutes), $5.00 (12 hours)$2$0.10$10
Over 272K input tokens$4$0.20$15
Price history
DateInputCachedOutput
openai.com · 2 · Corroborated$2$0.10$10
Over 272K input tokens$4$0.20$15

Context window

TokensContext windowInput limitOutput limit
Default1,050,000922,000128,000
Context history
DateContext windowInput limitOutput limit
As of developers.openai.com · 2 · Corroborated1,050,000922,000128,000

Benchmark results

BenchmarkHarness and effortScore
GPQA Diamondno toolsHarness not specifiedEffort not specified96.0%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Terminal-Bench 4.1Stanford University and Laude Institute’s leaderboard, 18 Nov 2026Codexmax effort61.5%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Terminal-Bench Science 0.1Harness not specifiedmax effort68.0%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Harness not specifiedlow effort43.4%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Harness not specifiedmedium effort63.1%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Harness not specifiedhigh effort67.7%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Harness not specifiedxhigh effort64.3%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
FrontierCode 2.0Score, all tasks (132)Run by CognitionHarness not specifiedmax effort43.8%Corroborated The same number is printed in 1 other document: 1
Run by CognitionHarness not specifiedlow effort33.1%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Run by CognitionHarness not specifiedmedium effort38.3%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Run by CognitionHarness not specifiedhigh effort39.7%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Run by CognitionHarness not specifiedxhigh effort39.8%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Cognition’s leaderboard, 18 Nov 2026 · in FrontierCode LeaderboardCodexmax effort43.8%Corroborated The same number is printed in 1 other document: 1
FrontierCode 2.0Pass rate, all tasks (132)Cognition’s leaderboard, 18 Nov 2026Codexmax effort48.2%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
SEC-bench Pro 260505pass@1Harness not specifiedEffort not specified88.34%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Agents' Last Exam v1.1Score, all tasks (122)Setting from Berkeley RDI’s leaderboardHarness not specifiedmax effort73.0%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Harness not specifiedlow effort66.7%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Harness not specifiedmedium effort70.3%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Harness not specifiedhigh effort72.0%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Harness not specifiedxhigh effort72.5%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Benchmark sources

Reported by OpenAI

Compare

Related

Sources

All models