WeirdMLLeaderboard
Solve an unusual machine-learning task end to end: write the training code, run it, and hit an accuracy target.
54.5%
58.7% of the best (92.9%)
#38 of 69
OpenAIUnited States· released Sep 2025
GPT-5 Codex
Provisional: 3 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.
Third-party publishedLow confidence (thin coverage)
Snapshot September 8, 2026
OpenCharts Index
53.0
Provisional: not enough comparable tests or counting categories to rank yet · 3 comparable tests · 0 counting categories
Drop any single test and the Index lands between 47.3 and 55.5.
Needs more results (1 of 8 so far)
Averaging 52.3 on the 1 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
GPT-5 Codex, in 3 results.
Its widest gap is on APEX-Agents, 57.6% behind the best published result. No comparable results yet in reasoning, knowledge, math, multimodal, human preference and long context. Between 40% and 50% behind the best published results, on average.
Strongest category
No category counts yet: a category counts once it has more than one comparable test and the model has taken at least half of them. The scores it has are below.
Best single result
WeirdML
Solve an unusual machine-learning task end to end: write the training code, run it, and hit an accuracy target.
54.5%
58.7% of the best (92.9%) · #38 of 69.
Real machine-learning work is messy and unfamiliar. This rewards models that can experiment rather than recite a tutorial.
Where it trails
APEX-Agents
Complete professional-services work in consulting, law and finance as an agent, graded against expert rubrics.
20.1%
57.6% behind the best published result (47.4%) · #29 of 49.
Expert-graded knowledge work is what professional teams would actually delegate.
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.
- Reasoning
- Knowledge
- Math
- Multimodal
- Human preference
- Long context
Covered, but not averaged into the Index:
- Coding1 of 7 tests · needs 4 · mean 58.7
- Agentic2 of 8 tests · needs 4 · mean 47.3
See it at work
What GPT-5 Codex was asked to do, and how it did.
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.
Coding · mean 58.7 across 1 comparable test · not counted toward the Index yet (fewer than half the category's tests)
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
GPT-5 Codex against whoever you pick.
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 Codex
provisional·Index 53.0C-
0 : 3
wins · 0 ties · 3 shared
GPT-5.5
#7·Index 80.2B+
GPT-5.5 comes out ahead on 3 of the 3 tests both have taken.
Category means
Reasoning
—89.4Knowledge
—83.3Coding
58.774.8Math
—74.9Agentic
47.381.5Multimodal
—92.3Human preference
—92.8
Shared tests · biggest gaps first
- Terminal-BenchAgentic−47.744.3%84.7%
- APEX-AgentsAgentic−38.820.1%38.5%
- WeirdMLCoding−32.754.5%84.9%
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-5.5Scores
Every test, every source
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
- Reasoning—
- Knowledge—
- Coding58.7 · 1/7
- Math—
- Agentic47.3 · 2/8
- Multimodal—
- Human preference—
- Long context—
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.
WeirdMLLeaderboardCoding
54.5%
58.7% of the best (92.9%)
46th percentile of 69
Terminal-BenchLeaderboardAgentic
44.3%
52.3% of the best (84.7%)
49th percentile of 40
agent: Codex CLI
Source: Epoch AI — AI Benchmarking HubAPEX-AgentsLeaderboardAgentic
20.1%
42.4% of the best (47.4%)
42th percentile of 49
high
Source: Epoch AI — AI Benchmarking HubTop of the leaderboard
Compare with its neighbours.
The best models, ranked here, working inside Theo.
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.