Turing Agent
An end-to-end formal verification agent framework for research-level mathematical problems that adopts a dual-scale architecture integrating theorem formalization and proof formalization agents, combined with mechanisms such as structured proof decomposition, parallel verification, and selective semantic review.
Timeline (last 90 days)
Turing Agent is an agent framework designed to orchestrate end-to-end formal verification, specifically targeting research-level mathematical problems.
Turing Agent achieved a faithful formalization success rate of about 57-62% on Formal TCS Bench.
Turing Agent achieved an overall pass@1 score of 83% on the TCS-focused RLMEval subset.
Turing Agent's tests used models such as GLM-5.3 and DeepSeek-V4-Pro.
Turing Agent, built on DeepSeek, successfully completed the formal verification of three previously unsolved TCS tasks from OpenAI's October 2026 collection.
One particularly difficult TCS problem required about one hundred formally verified auxiliary statements to complete.
All Facts (6)
Turing Agent is an agent framework designed to orchestrate end-to-end formal verification, specifically targeting research-level mathematical problems.
50%UnverifiedTuring Agent achieved a faithful formalization success rate of about 57-62% on Formal TCS Bench.
50%UnverifiedTuring Agent achieved an overall pass@1 score of 83% on the TCS-focused RLMEval subset.
50%UnverifiedTuring Agent's tests used models such as GLM-5.3 and DeepSeek-V4-Pro.
50%UnverifiedOne particularly difficult TCS problem required about one hundred formally verified auxiliary statements to complete.
50%UnverifiedTuring Agent, built on DeepSeek, successfully completed the formal verification of three previously unsolved TCS tasks from OpenAI's October 2026 collection.
50%