OpenAIUnited States· released Feb 2026
Provisional: 7 comparable tests so far. Ranked once it reaches 6 across 3 counting categories.
OpenAI, United States. One of 26 models from this lab on the Index.
Snapshot September 8, 2026
OpenCharts Index
80.9
Provisional: not enough comparable tests or counting categories to rank yet · 7 comparable tests · 1 counting categories
Drop any single test and the Index lands between 70.9 and 80.9.
Needs more results (3 of 8 so far)
Averaging 88.8 on the 3 it has; the score opens at 4 of 8.
Distance to today's frontier on the hardest tests. Not a measure of intelligence.
What it's known for
Strongest in agentic (80.9). Its widest gap is on Code Arena (WebDev), 80.7% behind the best published result. No comparable results yet in reasoning, knowledge, math, multimodal, human preference and long context. Between 15% and 20% behind the best published results, on average.
Strongest category
Agentic
Long-horizon tasks with tools: terminals, desktops, browsers, whole jobs.
80.9
Mean of 4 of the category's 8 comparable tests. Counts toward the Index.
Best single result
Complete a real task inside a terminal, such as setting up a service, fixing a build or wrangling data, verified by tests.
78.4%
92.6% of the best (84.7%) · #6 of 40.
The terminal is where agents do real operations work. This shows whether one can be left alone with a shell.
Where it trails
Build a web app from the same prompt as a rival model; real users vote blind on the result.
1,409
80.7 points short of parity with the board leader (1,797) · #53 of 86.
People judging finished apps side by side is the most honest measure of front-end quality.
Not measured yet
No comparable result yet in these 6 categories. A blank is a blank, never a zero, and it does not lower the Index.
Covered, but not averaged into the Index:
See it at work
Every test it has taken, by category, with the kind of task it faced, who produced the number and where the result landed against the best published one. The examples are original and illustrative, never items from the datasets themselves.
Agentic · mean 80.9 across 4 comparable tests · counts toward the Index
Complete a real task inside a terminal, such as setting up a service, fixing a build or wrangling data, verified by tests.
78.4%
92.6% of the best (84.7%)
#6 of 40
agent: SageAgent
Harnesses differ by model, so this compares model-plus-harness systems, not models alone.
Source: Terminal-BenchComplete software tasks of increasing length; the score is the task length a model finishes with 50% reliability.
5.8 h
84.2% of the best (17.4 h) on a log scale
#4 of 14
METR publishes wide confidence intervals around each horizon; the point estimate is used here.
Source: METRRun a simulated vending business for a year: ordering, pricing and cash flow. The score is the final balance.
$5,940
79.6% of the best ($11,182) on a log scale
#15 of 53
Andon Labs estimates a strong human operator at roughly $63,000, so every model is far from the ceiling.
Source: Andon LabsComplete professional-services work in consulting, law and finance as an agent, graded against expert rubrics.
31.8%
67.1% of the best (47.4%)
#20 of 49
Every result is read against the best published one on its test: percent scores as a share of the best above chance, Arena ratings as the expected win rate against the board leader, open-ended values on a log scale. Sitting a harder exam never lowers a model. Every number links to the publisher that produced it.
Head to head
Choose a rival. Every counted test both have taken appears side by side as a share of the best published result on that test, with the category means above. A win is a higher share; a gap under 2 points is a tie, because one item on a 45-problem exam is 2.2 points and publishers' own standard errors are of that order.
GPT-5.3 Codex
provisional·Index 80.9B+
0 : 3
wins · 0 ties · 3 shared
GPT-6 Astra
#1·Index 99.0A+
GPT-6 Astra comes out ahead on 3 of the 3 tests both have taken.
Category means
Reasoning
Knowledge
Coding
Math
Agentic
Shared tests · biggest gaps first
Bars are distances to the best published result on each test (percent scores above chance, Arena ratings as win rate against the leader, open-ended values on a log scale). Raw values as the publishers report them.
Open GPT-6 AstraScores
The best published run per benchmark, the raw value as the publisher reports it, who produced the number, how close it comes to the best published result on that test, and where the model sits among every model scored on it. Percent scores are read above chance, Arena ratings as a win rate against the board leader, open-ended values on a log scale. A test with too few models to compare against is shown but not counted.
Category means
Mean · tests taken of the category's comparable tests. A faded category either rests on fewer than half of them or is a single-test category (marked “beside”), which is shown next to the Index and never averaged into it.
74.8%
89.6% of the best (83.5%)
48th percentile of 26 · n = 500
high
Source: Epoch AI — AI Benchmarking Hub79.3%
85.4% of the best (92.9%)
87th percentile of 69
1,409
10% expected win rate against the board leader (1,797)
38th percentile of 86
codex-harness · board 2026-09-05
Source: Arena (LMArena) — Leaderboard Dataset78.4%
92.6% of the best (84.7%)
87th percentile of 40
agent: SageAgent
Source: Epoch AI — AI Benchmarking Hub5.8 h
84.2% of the best (17.4 h) on a log scale
77th percentile of 14
31.8%
67.1% of the best (47.4%)
60th percentile of 49
$5,940
79.6% of the best ($11,182) on a log scale
73th percentile of 53
Top of the leaderboard
28 of the models on this page run inside Theo today. Theo picks the right one for each step and always shows which engine answered.