Image said to be a slide from IPO roadshow materials spreads
An image said to be a slide from the materials for Anthropic's roadshow, the presentations to institutional investors ahead of its initial public offering (IPO), was posted on X on Nov. 13 (4:56:17 a.m. Nov. 14 Japan time) and spread. It was posted by Jacob Coxon (@hilbertspaess), whose profile reads "ai research." Coxon wrote that the slide was "a big factor in my decision to leave" and expressed surprise that Anthropic had made it public. The origin and authenticity of the material could not be verified, and Anthropic has not confirmed it. That Coxon worked at Anthropic has not been officially confirmed either.
This slide from the Anthropic IPO roadshow was a big factor in my decision to leave. It’s an insane plot and I’m surprised they made it public. It shows the compute efficiency of pretraining over time. The y-axis is a **log scale**: they are seeing continued superexponential algorithmic progress. Mythos was not a one-time deviation, but a point on a superexponential. They could already retrain Mythos with a fraction of the compute originally needed. It doesn’t strictly imply a software-only intelligence explosion- they have been spending more on R+D. But algorithmic improvements do not require ridiculous creative genius and are on the verge of being automated. You do the math.
Anthropic officially announced on Nov. 10 (the evening of the same day Japan time) that it had begun its roadshow for a Nasdaq listing (earlier article). However, the official announcement does not establish whether this slide is part of the material used in those presentations.
Chart has no numeric scale, and a footnote calls the placement illustrative
The attached image shows a chart titled "Pre-training compute multiplier over time (code)." The vertical axis reads "Compute multiplier (log scale)," but no numbers are marked on it. The horizontal axis is "Completion month," with ticks at July 2024, April 2025, January 2026 and October 2026. Three orange points sit near July 2024, near April 2025 and after January 2026, followed by an arrow extending up and to the right.
Beneath the chart, in small type, is a footnote: "The placement of the data points on the graph is based on internal approximations and is solely for illustrative purposes." Only the arrangement of the points and the slope can be read from the image; the values of the multiplier cannot.
Poster claims "superexponential" progress
In his post, Coxon claimed the chart shows the compute efficiency of pre-training over time and that algorithmic progress is continuing at a "superexponential" pace. Of "Claude Mythos," the model offered to vetted organizations (its latest version, 5.2, was officially released on Nov. 9), Coxon wrote that it was not a one-time deviation but a point on the curve, and that it could already be retrained with a fraction of the compute originally needed. These are Coxon's interpretations, and the figures cannot be checked against the chart's scale or Anthropic's materials.
In a follow-up post, Coxon said there are several reasons for not violating the non-disclosure agreement (NDA), and that it is not about being sued but about "the relationship of trust between Anthropic and its employees." Coxon added, "I really wish we could talk about the absolute numbers."
Researcher and former Claude Code staffer clash over the footnote
The image also drew pushback over the footnote. Omar Khattab (@lateinteraction), whose profile lists MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL), quoted the leak post at 2:19:15 p.m. Nov. 14 Japan time, noting that everyone was commenting on the "log scale" while a very small footnote says the placement of the points is "solely for illustrative purposes."
Thariq (@trq212) replied at 4:56:20 p.m. Japan time. Thariq's profile reads "formerly Claude Code @anthropicai," and a post on October 24 Japan time (October 23 Pacific Time) said Thariq had left Anthropic after more than a year working on Claude Code. Thariq asked why Khattab had cut the "based on internal approximations" part of the footnote with "[...]," and wrote, "you should take the graph and its implications seriously."
At 5:03:20 p.m. Japan time, Khattab responded that the phrase "solely for illustrative purposes" very literally makes the chart content-free. At 5:13:50 p.m., Thariq replied that the footnote is "obviously text that a legal department would require you to put out," but that the graph was not made to mislead.
