A solver-verified chinese logic benchmark found the best frontier model scores 37.5 percent on hard…
a solver-verified chinese logic benchmark found the best frontier model scores 37.5 percent on hard items, and even with formalization help the highest joint score is 60 percent. reasoning demos look sharp until they hit tests that machines can verify.
the gap between demoed reasoning and provable reasoning is still the real benchmark.
Context
The arXiv paper LLMEval-Logic (2605.19597), May 2026, introduces a solver-verified Chinese benchmark for logical reasoning with adversarial hardening. Items are forward-authored from realistic scenarios, gold natural-language to formal-language alignment is audited, answers are verified with the Z3 solver, and there are expert rubrics. It has paired Base and Hard subsets, where Hard items chain about five counterfactual sub-questions and count as correct only if every sub-question is correct, and two evaluation axes, one being formalization grading.
The benchmark's identity, Z3 verification and design are first-party. The 37.5 percent for the best model on hard items and the 60 percent highest joint score were not located in the text read, so the model, the axis and the meaning of joint are unverified. It is a May 2026 preprint, so the model set is as of then. The demos look sharp until line is the author's opinion.
Related work
- LLMEval-Logic (GitHub) ↗Snippet only, 12 May 2026.
Watch next
- The paper's results table.
Sources
- LLMEval-Logic (arXiv 2605.19597)arxiv.org
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 18 September 2026 at 16:36 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →