The record for this issue
Two claims overstated the work.
83 claims in the draft, each tested by a second agent against the page it cites. 78 held; 5 did not and were fixed or cut before this went out.
-
01
titleconfirmed
Bigger than the work.
The page says “Framing only, no checkable fact; consistent with the two overstated corrections below”
-
02
dekwrong
Quoted phrase: "the Riemann hypothesis, cracked."
The page says “Among the problems apparently cracked is the Riemann hypothesis, which concerns the distribution of prime numbers”
-
03
dekwrong
Quoted phrase: Adobe's "entire suite."
The page says “Developer says 'software is over' after using AI to recreate Adobe's entire Creative Cloud suite for free”
-
04
dekconfirmed
An alpha app (PhotoCraft) is what the headline called the entire suite
The page says “Stage: alpha”
-
05
dekconfirmed
A result about where zeta's zeros aren't (zero-free region)
The page says “Proves that every Dirichlet L-function, including ζ(s), is zero-free in Re s>7/8, resolving the quasi-Riemann hypothesis”
-
06
dekconfirmed
Anthropic's 75% price cut
The page says “On average, it now costs around 75% less to run”
-
07
dekconfirmed
The 75% holds if your prompts are short
The page says “priced 90% lower than Claude Haiku 4.5 for requests up to 100,000 tokens ... 50% lower for larger requests”
-
08
summaryconfirmed
Three AI claims from the past week (within 8-day window to Oct. 14)
The page says “Ynetnews dated 2026-10-07; TweakTown Oct. 8, 2026; Anthropic October 7, 2026”
-
09
summaryconfirmed
Topics: OpenAI's math manuscripts, a free Adobe-style app suite, Claude Haiku 5.5's price cut
The page says “The apps are free, distributed under MIT and Apache 2.0 licences”
-
10
art svg / art altconfirmed
Art: small box, pink speech bubble marked SOLVED; alt text matches
The page says “SVG contains a rect box with lines, a pink (#FFA3D1) speech-bubble path and text SOLVED; alt describes exactly this. Illustrates the headline's framing, not a factual claim”
-
11
The storyconfirmed
Claims checked against OpenAI's catalogue, an open-source project's roadmap and Anthropic's price table
The page says “Sources cited are openai/math CONTENTS/README, PhotoCraft ROADMAP.md and Anthropic's pricing table”
-
12
The storyconfirmed
Two made unfinished work sound finished; the third held up depending on what you send
The page says “Stage: alpha (PhotoCraft); README: results at different stages of verification; Anthropic footnote: 90%/50% tiers”
-
13
The correctionconfirmed
Ynetnews said OpenAI's model apparently cracked the Riemann hypothesis
The page says “Among the problems apparently cracked is the Riemann hypothesis”
-
14
The correctionconfirmed
...among solutions to 377 unsolved math problems
The page says “OpenAI unveils solutions to 377 unsolved math problems in major AI breakthrough”
- 15 confirmed
-
16
The correctionconfirmed
OpenAI posted 722 manuscripts
The page says “published 722 preprints in a GitHub repository”
-
17
The correctionconfirmed
Posted on Oct. 6
The page says “On October 6, OpenAI published 722 manuscripts (OpenAI post dateline October 6, 2026)”
-
18
The correctionconfirmed
Grouped into 372 families of results
The page says “372 families of related results spread across 17 fields”
-
19
The correctionconfirmed
Unreleased model
The page says “using an unreleased internal OpenAI model”
-
20
The correctionconfirmed
Model was given about 4,000 open problems
The page says “roughly 4,000 open problems (README: the model was posed approximately 4,000 problems)”
-
21
The correctionconfirmed
None proves the Riemann hypothesis
The page says “The 7/8 result is not the Riemann Hypothesis. (Catalogue's closest entry is 003, a zero-free half-plane.)”
-
22
The correctionconfirmed
RH: every nontrivial zero of zeta has real part 1/2
The page says “all the 'non-obvious' zeros of the zeta function are complex numbers with real part 1/2”
-
23
The correctionconfirmed
Closest catalogue entry claims no zeros with real part above 7/8
The page says “every Dirichlet L-function, including ζ(s), is zero-free in Re s>7/8”
-
24
The correctionconfirmed
A weaker statement OpenAI calls the quasi-Riemann hypothesis
The page says “resolving the quasi-Riemann hypothesis”
-
25
The correctionconfirmed
OpenAI says results are at different stages of verification
The page says “This collection includes results at different stages of verification”
-
26
The correctionconfirmed
On Oct. 7 OpenAI withdrew three manuscripts over a sign error
The page says “a sign error invalidates a stabilization-trace cancellation argument (three manuscripts withdrawn; page dated October 7, 2026). RW: on Oct. 7, OpenAI announced the withdrawal”
-
27
The correctionconfirmed
...and revised 14 others
The page says “We have revised 14 other manuscripts with proof repairs”
-
28
The correctionconfirmed
Checking has started
The page says “has withdrawn three of the manuscripts for errors”
-
29
The correctionconfirmed
Verdict: Wrong
The page says “The 7/8 result is not the Riemann Hypothesis”
-
30
The correctionconfirmed
Near the famous problem, not on it
The page says “zero-free in Re s>7/8”
-
31
The correctionconfirmed
Geometry and label: axis 0 at x=80, 1 at x=360, dashed line at x=220 (=1/2), pink band from x=325 (=7/8) labelled 7/8
The page says “80+0.875*280=325; 80+0.5*280=220. Zero-free in Re s>7/8”
-
32
The correctionconfirmed
Alt text describes axis, dashed middle line with circles, pink band at 7/8
The page says “Matches SVG elements”
-
33
The correctionconfirmed
RH puts every nontrivial zero on the line at real part 1/2
The page says “with real part 1/2”
-
34
The correctionconfirmed
OpenAI's entry claims the band beyond 7/8 has none; says nothing about the space between
The page says “zero-free in Re s>7/8”
-
35
The correctionconfirmed
A developer used AI to recreate Adobe's entire Creative Cloud suite, for free
The page says “using AI to recreate Adobe's entire Creative Cloud suite for free”
-
36
The correctionconfirmed
TweakTown headline, Oct. 8
The page says “October 8, 2026, 3:30 PM CDT”
-
37
The correctionconfirmed
The ArtCraft project released seven apps
The page says “has released seven open-source applications designed to recreate Adobe's Creative Cloud suite; caption: seven ArtCraft creative tools”
-
38
The correctionconfirmed
Free and open-source
The page says “The apps are free, distributed under MIT and Apache 2.0 licences”
-
39
The correctionconfirmed
Modeled on Adobe's apps
The page says “PhotoCraft for Photoshop, VectorCraft for Illustrator, FilmCraft for Premiere Pro”
-
40
The correctionconfirmed
Built with Claude Opus 5.5, TweakTown reports
The page says “all seven were built using Anthropic's Claude Opus 5.5 and written in Rust”
-
41
The correctionconfirmed
PhotoCraft is the Photoshop stand-in
The page says “PhotoCraft for Photoshop (TweakTown does not link the storytold/photocraft repo; the roadmap measures itself against Photoshop 2026 and uses Opus 5.5 agent hours, consistent with it being the same app)”
-
42
The correctionconfirmed
Roadmap last updated Oct. 11
The page says “Last updated: 2026-10-11”
- 43 confirmed
-
44
The correctionconfirmed
Oct. 10 re-measure: feature breadth about 76% of Photoshop
The page says “2026-10-10: Full re-measure against Photoshop 2026 27.11.0 ...: breadth ~76%, ready ~45%, alpha”
- 45 confirmed
-
46
The correctionwrong
The Oct. 10 re-measure included 56 of Photoshop's 68 tools
The page says “Revision history 2026-10-11: Toolbox count 56 of 68 (12 missing) with the Convert Point tool (#3204)”
-
47
The correctionconfirmed
Beta needs about 75%
The page says “Beta needs ~75% ready and no blocking PSD gap”
-
48
The correctionconfirmed
First commit on Sept. 30
The page says “2026-09-30: First commit”
-
49
The correctionconfirmed
Under two weeks from first commit to that point
The page says “2026-09-30 first commit to 2026-10-10 re-measure = 10 days”
-
50
The correctionconfirmed
The project rates itself honestly; the headline didn't
The page says “Honest assessment: ready for real work 25–35% (Oct. 5); Stage: alpha”
-
51
The correctionconfirmed
Verdict: Overstated
The page says “breadth ~76%, ready ~45%, alpha”
-
52
The correctionconfirmed
Wide, not yet deep
The page says “breadth ~76%, ready ~45%”
-
53
The correctionconfirmed
PhotoCraft against Photoshop, the project's own Oct. 10 re-measure
The page says “2026-10-10: Full re-measure against Photoshop 2026”
- 54 confirmed
- 55 confirmed
-
56
The correctionconfirmed
75%
The page says “Beta needs ~75% ready”
-
57
The correctionconfirmed
The project's own estimates
The page says “estimated: weighted by dimension below”
-
58
The correctionconfirmed
Readiness is its weighted score across several areas
The page says “Ready for real work: estimated: weighted by dimension below (10 dimensions listed)”
- 59 unsupported
-
60
The correctionconfirmed
Haiku 5.5 costs around 75% less to run than Haiku 4.5
The page says “On average, it now costs around 75% less to run”
-
61
The correctionconfirmed
Anthropic announcement, Oct. 7
The page says “October 7, 2026”
-
62
The correctionconfirmed
90% cheaper per token for prompts up to 100,000 tokens
The page says “priced 90% lower than Claude Haiku 4.5 for requests up to 100,000 tokens”
-
63
The correctionconfirmed
$0.10 per million input tokens against $1
The page says “Input: $0.10 / $0.50 (Haiku 5.5); Haiku 4.5 Input: $1.00”
-
64
The correctionconfirmed
50% cheaper above that ($0.50)
The page says “$0.50 / MTok for prompts over 100,000 tokens”
-
65
The correctionconfirmed
Anthropic's 75% is a blend
The page says “On average ... footnote 2: 90% lower ... 50% lower for larger requests”
-
66
The correctionconfirmed
About 90% of Haiku 4.5 requests were short
The page says “On Haiku 4.5, 90% of requests fell into the former category”
-
67
The correctionconfirmed
The figure allows for the new tokenizer
The page says “footnote: calculation accounts for the tokenizer change; Haiku 5.5 has an updated tokenizer”
-
68
The correctionconfirmed
Docs: same text counts as about 30% more tokens than Haiku 4.5
The page says “the same text counts as approximately 30 percent more tokens than on Claude Haiku 4.5”
-
69
The correctionconfirmed
Short request costs about 87% less after 30% more tokens
The page says “Arithmetic: 1.3 x $0.10 = $0.13 vs $1.00 = 87% less”
-
70
The correctionconfirmed
Long request costs about 35% less
The page says “Arithmetic: 1.3 x $0.50 = $0.65 vs $1.00 = 35% less”
-
71
The correctionunsupported
Batch requests take a further 50% off either tier, per Anthropic's documentation
-
72
The correctionconfirmed
Verdict: Right, but
The page says “On average ... around 75% less; tiers 90% / 50%”
-
73
The correctionconfirmed
The cut shrinks for long prompts
The page says “90% lower ... 50% lower for larger requests”
-
74
The correctionconfirmed
Haiku 5.5 against Haiku 4.5, input tokens
The page says “Input $0.10/$0.50 vs $1.00”
- 75 confirmed
-
76
The correctionconfirmed
50%
The page says “50% lower for larger requests”
-
77
The correctionconfirmed
87% (our arithmetic)
The page says “1.3 x 0.10 = 0.13 vs 1.00”
-
78
The correctionconfirmed
35% (our arithmetic)
The page says “1.3 x 0.50 = 0.65 vs 1.00”
-
79
The correctionconfirmed
Output prices fall by the same shares
The page says “Output: $0.50 / $2.50 vs Haiku 4.5 $5.00 = 90% / 50%”
-
80
The correctionconfirmed
Anthropic's 75% average assumes most requests are short
The page says “On Haiku 4.5, 90% of requests fell into the former category”
-
81
The checklist, item 1confirmed
Name the exact result: a famous problem in a headline may stand in for a weaker one nearby
The page says “Generic habit; matches Correction 1 (quasi-Riemann entry vs RH headline)”
-
82
The checklist, item 2confirmed
Readmes and roadmaps say alpha when headlines don't
The page says “Stage: alpha (vs TweakTown headline); general habit”
-
83
The checklist, item 3confirmed
An average cut is computed from someone else's traffic
The page says “On Haiku 4.5, 90% of requests fell into the former category”
The checker's notes
- Window: every cited fact relied on is dated on or before 2026-10-14 (Ynetnews Oct 7, TweakTown Oct 8, Anthropic Oct 7, OpenAI Oct 6, LDS Oct 7, Retraction Watch Oct 8, history.md Oct 7, roadmap last updated Oct 11). OpenAI's post carries an Oct 15 update and the repo README now shows 719 manuscripts (post-withdrawal); the issue relies on neither. No ahead/what-to-watch section.
- No compliance issues: AI product pricing only, no securities or market content.
- No quotes over 14 words. The two dek quotes are flagged WRONG because they are not verbatim.