The claim traces to a real, identifiable primary artifact published the day before the Instagram post. Anthropic published "Claude-shaped science" on October 1, 2026, a guest post by theoretical physicist Matthew Schwartz describing BootLoops, an open-source toolkit he built with Claude for exact calculations in quantitative science, released publicly the same day under the MIT License.

The specific numbers in the claim appear in the post. The decisive sentence, quoted verbatim, reports "15 reproductions of known results by this new method and 15 that had never before been computed" , out of 30 integrals carried end to end. This is corroborated by independent paraphrases: Schwartz reports the toolkit ultimately computed 30 integrals end to end, 15 reproductions of known results by the new method and 15, including elliptic Feynman integrals, that had never before been computed.

The post itself scopes the novelty carefully. Schwartz writes that elliptic integrals are hard even for specialists, that only a handful of elliptic Feynman integrals have ever been computed, and "none completely by the bootstrap, at least to my or Claude's knowledge," and that the issue is not that the methods would not work but that nobody had tried .

The "20 minutes" detail in the caption is also in the source. He reports that Claude reproduced the results of one of his papers in about 20 minutes, against the weeks he had spent writing his own code, and that it identified a better algorithm he was unaware of. The scale figures check out as well: Schwartz reports the effort produced 36 manuscripts across 18 fields with 19 coauthors over three months, drawn from about 400 candidate problems; the project site states BootLoops 1.0 applications spanned twenty-two fields.

Model attribution is more layered than the claim suggests. BootLoops grew out of an iterative process with Claude that began in December 2025, Schwartz started with Claude Opus 4.5, and by the summer of 2026 the workflow had moved to Claude Fable 5. A separate account states he began the project when Anthropic released Claude Fable 5 in Summer 2026, after an earlier experiment described in his post Vibe Physics in which he used Claude Opus 4.5 as a research assistant in December 2025. One roundup describes the harness as running Claude "(mostly Fable 5)" . The toolkit itself is not Claude-specific: despite its origins, BootLoops is model-agnostic , and the harness is model independent and can be driven by Claude, Gemini or ChatGPT .

"Claude Fable 5" is a real Anthropic model, not a garbled name. Anthropic released Claude Fable 5, its first Mythos-class model available to the public, with guardrails blocking responses in high-risk areas like cybersecurity and biology. Official docs describe it as Anthropic's most capable widely released model, built for demanding reasoning and long-horizon agentic work .

The verification status of the results is the material gap. The results are the author's own claims, most projects are still undergoing further verification, and Schwartz has been a visiting researcher at Anthropic; BootLoops is not an Anthropic project. Anthropic's own page carries that disclosure: during this project Schwartz has been working as a visiting researcher at Anthropic, and BootLoops is owned and maintained by Matthew Schwartz rather than being an Anthropic project. One roundup states plainly that results are self-reported and mostly not yet peer-reviewed .

The post also records limits, consistent with the caption's caveats. His mitigations included encoding protocol skills in the harness, insisting on seeing plots himself, and rechecking results as an adversarial referee.

Schwartz lists failure modes including premature victory claims and poor time estimates, and discloses he is a visiting researcher at Anthropic. His own summary of the state of play is "Claude and GPT are good at science, but they are not scientists."