“GPT-5 is barely better than last month’s flavor of the month (Grok 4); on some metrics (ARC-AGI-2) it’s actually worse.”
Our reading, the sentence that was checked:
GPT-5 is barely better than Grok 4 and worse on some metrics such as ARC-AGI-2.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, GPT-5: Overdue, overhyped and underwhelming. And that’s not the worst of it., paragraph 28.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-7768d1-a3
We recorded 31 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 2025-08-09. 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
- GPT-5 is barely better than
- Grok 4
- on some metrics (ARC-AGI-2) it's actually worse
Our interpretation
- Read as two parts: (a) on broad benchmark composites at launch, GPT-5 led Grok 4 by only a small margin; (b) on ARC-AGI-2, GPT-5 scored below Grok 4.
- Use the launch-day scores of the flagship settings (GPT-5 high vs Grok 4 / Grok 4 Thinking). Target period is August 2025.
Counts as supported if: Launch benchmark composites show GPT-5 only slightly ahead of Grok 4, and ARC-AGI-2 shows GPT-5 below Grok 4.
Counts as contradicted if: GPT-5 leads Grok 4 by a wide margin on broad composites, or GPT-5 scores at or above Grok 4 on ARC-AGI-2 at launch.
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 claim has two parts, and reports dated 7 August 2025 support both. Composite: OfficeChai reported that GPT-5 at its 'max' reasoning setting scored 69 on Artificial Analysis's Intelligence Index, against 68 for Grok 4 (e3). That is a one-point lead, which meets the contract's condition of GPT-5 being only slightly ahead. The same report quotes Artificial Analysis giving GPT-5 a score of 68 at High reasoning effort (e3). The report therefore does not settle the exact margin, but neither figure shows a wide lead. ARC-AGI-2: The Decoder reported that ARC Prize scored GPT-5 (High) at 9.9% (e1) and Grok 4 (Thinking) at about 16% (e2). On that benchmark GPT-5 scored below Grok 4. The same report gives GPT-5's cost as $0.73 per task and Grok 4's as $2 to $4 (e1, e2). The contract compares scores, not cost.
Limitations
- The scores come from news reports that cite ARC Prize and Artificial Analysis. The primary leaderboards were not read directly.
- The composite report gives two GPT-5 scores, 69 at 'max' and 68 at High reasoning effort, so the exact margin over Grok 4's 68 is uncertain.
- Other leaderboards, such as LMArena, ranked GPT-5 first (e3). 'Barely better' is judged only on the Artificial Analysis composite index named in the contract.
- The evidence covers launch-era scores in August 2025 only, not later model versions or later results.
Evidence (3 items)
Supports the claim: The Decoder,
GPT-5 (High) scored 9.9 percent at a cost of $0.73 per task, according to ARC Prize.
Why it matters here: Gives GPT-5's launch ARC-AGI-2 score as reported by ARC Prize.
- Where
- paragraph 1
- Publisher
- The Decoder
- Kind of source
- News report
- Describes events of
- Public from
Supports the claim: The Decoder,
Grok 4 (Thinking) did better on ARC-AGI-2 at roughly 16 percent, but at a much higher $2 to $4 per task.
Why it matters here: Grok 4 scored about 16% against GPT-5's 9.9%, so GPT-5 was worse on ARC-AGI-2, though cheaper per task.
- Where
- paragraph 2
- Publisher
- The Decoder
- Kind of source
- News report
- Describes events of
- Public from
Supports the claim: OfficeChai,
GPT-5 with ‘max’ reasoning scored 69. This is higher than the 68 Grok 4 had managed
Why it matters here: On a broad composite of eight benchmarks, GPT-5 led Grok 4 by one point, 69 to 68. That matches 'barely better'.
- Where
- paragraph 2
- Publisher
- OfficeChai
- Kind of source
- News report
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, GPT-5: Overdue, overhyped and underwhelming. And that’s not the worst of it.
GPT-5 is barely better than last month’s flavor of the month (Grok 4); on some metrics (ARC-AGI-2) it’s actually worse.
- Where
- paragraph 28
- Published
- 2025-08-09
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-7768d1-a3: 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-gpt-5-overdue-overhyped-and-underwhelming-7768d1-a3.
HeldBy this site
The final-text check failed: rationale_supported_by_evidence; overclaims: Both parts hold for launch week in August 2025. | Either way, the gap is small, which fits 'barely better'..
About assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-7768d1-a2, which was not published.