# Meta and NYU’s h1 system beats elaborate theorem-proving workflows at lower cost > Source: > Published: 2026-11-01 19:30:48+00:00 # Meta and NYU’s h1 system beats elaborate theorem-proving workflows at lower cost Meta Platforms logo (public domain) via Wikimedia Commons A minimal agentic loop built on Lean feedback posts top benchmark scores and solves eight open problems Researchers at [Meta](https://cryptobriefing.com/markets/meta/) and New York University have built a stripped-down AI system for proving math theorems. It matched or beat carefully engineered Lean proof workflows, and it did so for less money. The system is called h1. Its core argument is that simplicity can beat complexity in automated theorem proving, provided the AI gets good feedback. ## What h1 actually does The work comes from Meta’s FAIR research team and NYU. Their paper is titled “h1: A Taste of the Bitter Lesson in Automated Theorem Proving.” It was released on October 25, 2026. The researchers describe h1 as a minimal agentic loop. It connects frontier large language models to Lean 4 and feeds them verifier results in real time. Most automated provers wrap the language model in an elaborate scaffold. They use predefined workflows that split problems into pieces, route subtasks, and orchestrate many moving parts. The h1 framework skips that machinery. Instead, it leans on a few simple mechanisms: - **Persistent file editing** , so the model works on a proof file that keeps its state between steps - **Immediate compiler feedback after every edit** , so errors surface right away - **Reflection** , using the language model’s own ability to reason about what went wrong The paper argues that h1’s performance comes mainly from an efficient interface with Lean. Fancy orchestration is not doing the heavy lifting. ## The benchmark numbers According to the research, h1 reached state-of-the-art performance on several formal theorem-proving benchmarks. ### AI, tech, and the markets they move—in one daily briefing. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. On PutnamBench, it scored 670 out of 672 using GPT-5.5. On the FATE-X assessment, h1 scored 90 out of 100. The system also resolved eight open problems in the FormalConjectures collection. These include variants of Erdős problems, a set of questions named after the prolific mathematician Paul Erdős. The most direct comparison is with DeepMind’s AlphaProof [Nexus](https://cryptobriefing.com/markets/nexus-4/). On Erdős problem resolutions, h1 posted an 82.5% success rate. AlphaProof Nexus came in at 76%. The cost gap is wider than the accuracy gap. h1 spent $121 per successful proof. AlphaProof Nexus spent $317. ## The bitter lesson, applied to math The paper’s title points to the “bitter lesson.” This is the idea that scalable, general-purpose methods powered by heavy computation tend to outperform specialized, handcrafted techniques over time. This is not Meta’s first attempt at the problem. The research builds on earlier Meta work, including the HTPS neural theorem prover introduced in 2022. The new paper reflects a shift in approach. Earlier systems focused on building smarter search around the model, while h1 focuses on letting the model interact directly with the verifier. Timing matters too. h1’s results depend on frontier models such as GPT-5.5, and the strong PutnamBench score was achieved with that model. ## What this means The cost figures are the most practical part. A drop from $317 to $121 per successful proof changes which problems are worth trying at all. When each attempt costs less, teams can run more attempts on harder or more speculative conjectures. Some caveats apply. Benchmark scores show performance on specific test sets, and PutnamBench is close to saturated at 670 out of 672. The h1 approach also depends heavily on the frontier model underneath it. **Disclosure:** This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our [Editorial Policy](https://cryptobriefing.com/editorial-policy/). ## More in AI [See all](https://cryptobriefing.com/artificial-intelligence/)