“(is OpenAI’s o1 100 times better than GPT4? Certainly not by their own data)”
Our reading, the sentence that was checked:
By OpenAI's own data, o1 is certainly not 100 times better than GPT-4.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, An epidemic of weirdly precise AI butt facts, paragraph 5.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-an-epidemic-of-weirdly-precise-ai-6dd8f9-a2
We recorded 14 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 2024-10-30. 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)
Conditions
- by OpenAI's own data (qualifier, stated by the speaker)
From the speaker
- is OpenAI's o1 100 times better than GPT4?
- Certainly not
- by their own data
Our interpretation
- Plain reading: none of OpenAI's own published o1 evaluations show anything close to a hundredfold gain over GPT-4-class models.
- OpenAI's launch post compares o1 with GPT-4o, its GPT-4-family model at the time. I treat that as the 'GPT4' comparison OpenAI's own data offers.
- The period is the state of OpenAI's published o1 data at the post date: from o1's announcement on 2024-09-12 to 2024-10-30.
Counts as supported if: OpenAI's published o1 results show gains well below 100x over GPT-4/GPT-4o on its reported metrics.
Counts as contradicted if: An OpenAI-published metric shows o1 roughly 100 times better than GPT-4 or GPT-4o.
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
OpenAI's o1 launch post, published on 12 September 2024, compares o1 with GPT-4o. On the 2024 AIME, GPT-4o solved about 12% of problems and o1 about 74% with a single sample, roughly 6x (e1). In the post's Appendix A table (e2), o1 scores 78.0 on GPQA Diamond against GPT-4o's 56.1, about 1.4x. MATH goes from 60.3 to 94.8 and MMLU from 88.0 to 90.8. The largest ratios in that table are about 8x, for AIME pass@1 (9.3 to 74.4) and the CodeForces percentile (11.0 to 89.0). None of the scores in the launch post comes near 100x. Marcus's statement that OpenAI's own data does not show o1 as 100 times better than GPT-4 holds for these published results.
Limitations
- OpenAI's baseline is GPT-4o, not the original GPT-4.
- Only the launch post was checked, not the o1 system card or later OpenAI publications.
- 'Better' is not defined. Ratios of benchmark scores are one reasonable measure, and scores with a fixed maximum limit how large a ratio can be.
Evidence (2 items)
Supports the claim: OpenAI,
GPT‑4o only solved on average 12% (1.8/15) of problems. o1 averaged 74% (11.1/15) with a single sample per problem
Why it matters here: On AIME, one of OpenAI's best showcases for o1, the gain is about sixfold (74% vs 12%), far below 100x.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Arithmetic
- 74 / 12 (inputs: o1 AIME single-sample 74%; GPT-4o AIME 12%; about 6.2x)
- Describes events of
- Public from
Supports the claim: OpenAI,
GPQA Diamond cons@64 56.1 78.3 78.0
Why it matters here: On GPQA Diamond, GPT-4o scored 56.1 and o1 78.0, a gain of about 1.4x. Other rows (MMLU 88.0 vs 90.8, MATH 60.3 vs 94.8) also show modest gains.
- Where
- Appendix A table
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Arithmetic
- 78.0 / 56.1 (inputs: o1 GPQA Diamond 78.0; GPT-4o GPQA Diamond 56.1; about 1.4x)
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, An epidemic of weirdly precise AI butt facts
(is OpenAI’s o1 100 times better than GPT4? Certainly not by their own data)
- Where
- paragraph 5
- Published
- 2024-10-30
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-an-epidemic-of-weirdly-precise-ai-6dd8f9-a2: 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-an-epidemic-of-weirdly-precise-ai-6dd8f9-a2.