Fetching from the wire…
Public story · 2026-07-20 · high
Alpöge and a collaborator used Fable 5 to find a degree-7 counterexample where mathematicians expected degree 200.
Why now: One day after Alpöge's post, the dispute over how much of the search Fable 5 actually drove is still unresolved.
Levent Alpöge, a mathematician at Anthropic, posted on July 19 that he and a collaborator found a counterexample to the Jacobian Conjecture. The conjecture has been open since 1939.
That matters for mathematicians trying to figure out whether AI can do real research work or just win benchmarks. Most "AI does math" claims land as a score or a proof too dense for outsiders to check. This one is a specific object: a degree-7 counterexample in three variables, verifiable by hand instead of trusted on faith.
Mathematicians had estimated any counterexample would need degree 200 or higher. Alpöge and his collaborator, working with Claude Fable 5, landed on degree 7. The Hacker News thread built around the post moved fast, and most of the argument wasn't over whether the math checks out. It was over how much of the search Fable 5 actually drove versus how much the two mathematicians steered by hand.
That's the right question to ask about any AI math claim, and it's probably not one outsiders can answer. Nothing in the thread settles it either way.
The hand-checkable part is the real claim here, and it's testable. If independent mathematicians outside Anthropic don't publish a check of the degree-7 counterexample soon, that's telling. It would mean the "verify it yourself" pitch did more work than actual verification.
Each link below shares sources, entities, or timing with this story.
Moonshot AI released Kimi K3, a sparse mixture-of-experts activating 16 of 896 experts per token. That's about 1.8% of the pool live at any moment, with a 1M-token context window and native vision. Two new architectural pieces show up: Kimi Delta Attention and Attention Residu...
Anthropic commissioned the independent evaluator to test 72 injection scenarios, held out from Anthropic, each run 10 times against Fable 5, Opus 5, and Sonnet 5 as of July 17. Clean sweep. TechCrunch has the details. A third-party held-out eval is a much stronger claim than i...
After a June 12 US export-control order pulled Fable 5 and Mythos 5 offline for nearly three weeks, Anthropic brought Fable 5 back once the order lifted. The order followed an Amazon report that Fable 5's safeguards could be bypassed to surface software vulnerabilities. Anthro...
43.3% on Frontier-Bench v0.1. Opus 4.8 scored 18.7%. That's not an incremental bump, that's the same benchmark with a different shape of answer. Anthropic released Claude Opus 5 on July 24 at $5/$25 per million input/output tokens, exactly half of Fable 5's $10/$50, while matc...
Anthropic published "Redeploying Fable 5" on July 18, and the headline reads like good news until you get to the metering. Fable 5 becomes a permanent subscription feature for Max and Team Premium. Good. But it's metered at 50% of standard weekly limits, meaning every Fable to...
Anthropic shipped Fable 5 on June 9. Willison spent ~5.5 hours stress-testing it: slow and expensive, but it handled everything he threw at it, including agentic coding. (Simon Willison) The tell that it's a real working model and not a benchmark queen: because it post-dated A...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.