“the crown jewel that they reported on the demo — the 75% on Francois Chollet’s ARC test (once called ARC-AGI) doesn’t readily replicate.”
Our reading, the sentence that was checked:
OpenAI's reported 75% o3 score on ARC-AGI does not readily replicate.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, OpenAI’s dirty December o3 demo doesn’t readily replicate, paragraph 3.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-openais-dirty-december-o3-demo-doesnt-2f37cd-a1
We recorded 8 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2025-04-23. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
From the speaker
- the 75%
- Francois Chollet’s ARC test
- doesn’t readily replicate
- the model that is released as “o3 (high)”
Our interpretation
- The surrounding text makes this a claim about practical reproduction using publicly released April 2025 o3.
- The target is the latest public release-testing period before the post, April 16–23, 2025.
- Readily replicate means completing a credible independent evaluation near the advertised score without exceptional completion failures.
- This does not mean the original December preview score was fabricated or that the identical preview configuration failed a repeat experiment.
Counts as supported if: Independent testing of released o3 on the corresponding ARC-AGI-1 evaluation reports substantially lower complete-run scores or completion failures that prevent reliable reproduction near 75%.
Counts as contradicted if: Independent testing during the target period routinely completes comparable evaluations of released o3 at approximately 75% or better.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Supported
The dated evidence matches what the claim committed to, under the criteria frozen before grading.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The April evaluator could not obtain reliable high-reasoning coverage, and its reported medium-setting score was 53%, well below the December result. This supports the practical claim about released o3. It does not show that the original experiment failed replication: ARC Prize explicitly says the production model and available compute differed and continues to affirm the December score. The report's 53% also differs from the 56% mentioned elsewhere in Marcus's post.
Limitations
- Support depends on the released-model reading supplied by the surrounding text.
- High reasoning's 82% accuracy covered only returned tasks and cannot establish full-set success.
- April 22 dates the documented incomplete evaluation state; individual requests may have run earlier.
Evidence (1 item)
Supports the claim: ARC Prize,
did not return enough task completions to support reliable scoring.
Why it matters here: As reported April 22, o3-medium scored 53% on ARC-AGI-1's semi-private set, while o3-high returned only 37 of 100 tasks; the evaluator rejected its 82% returned-task accuracy as unrepresentative. The same report affirms the December result but explains that production o3 differed in model, training and available compute, supporting practical non-reproduction rather than invalidity of the original experiment.
- Publisher
- ARC Prize
- Kind of source
- Independent measurement
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, OpenAI’s dirty December o3 demo doesn’t readily replicate
the crown jewel that they reported on the demo — the 75% on Francois Chollet’s ARC test (once called ARC-AGI) doesn’t readily replicate.
- Where
- paragraph 3
- Published
- 2025-04-23
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-openais-dirty-december-o3-demo-doesnt-2f37cd-a1: Current. Outcome: Supported. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.
About assessment marcus-openais-dirty-december-o3-demo-doesnt-2f37cd-a1.