Yuji Tachikawa, a University of Tokyo professor, says Claude Fable 5 solved a quantum field theory problem stalled for six months. The model found a calculation error, hit the same dead end as researchers, then expanded a new approach after one follow-up prompt.
Key facts
- Problem stalled for six months before Claude Fable 5 intervention.
- Model found a calculation error in existing research notes.
- Claude Fable 5 used SymPy to verify its own predictions.
- Tachikawa called the observation 'non-trivial' and essentially solved.
- Account is single researcher's report, not peer-reviewed.
Yuji Tachikawa, a University of Tokyo professor working on quantum field theory and string theory, posted on X that Anthropic's Claude Fable 5 unlocked a collaborative research problem that had seen no progress for six months. According to @kimmonismus, Tachikawa gave the model his research notes on a whim. Fable found a calculation error, reached the same dead end as the researchers, and then expanded another approach after a follow-up prompt.
Tachikawa's verdict: “It made a non-trivial observation and essentially solved it.” The model also used SymPy to write code and verify its own predictions. His conclusion: “Fable probably seems like it properly understands string theory and has intuition too.”
This is one researcher's account, not a published or independently verified result. But it describes something far more consequential than another benchmark gain: a frontier model contributing a novel step to active theoretical physics research. The event echoes recent findings where AI models have proposed conjectures in knot theory and discovered new mathematical structures, but this is among the first public accounts of a model directly advancing work in string theory — a domain requiring deep intuition about symmetries, dualities, and high-dimensional geometry.
The incident raises questions about how frontier models like Claude Fable 5, which Anthropic has not yet formally released or benchmarked on physics reasoning tasks, achieve this capability. Tachikawa's report suggests the model may be combining formal mathematical reasoning with pattern recognition in ways that go beyond standard LLM capabilities. Anthropic has not commented on the report.
Key Takeaways
- Claude Fable 5 solved a string theory problem stalled for six months.
- Professor Yuji Tachikawa says the model made a non-trivial observation and used SymPy for verification.
What to watch
Watch for Anthropic to release a technical report or benchmark on Claude Fable 5's formal reasoning capabilities. Also track whether Tachikawa publishes a paper incorporating the model's contribution, which would provide independent verification.
