Research map / Large language models / Agents, code and math
FATE: A Formal Benchmark Series for Frontier Algebra of Multiple Difficulty Levels
Jiedong Jiang, Wanyi He, Wang Yuefeng, Guoxiong Gao, Yongle Hu, Jingting Wang, Nailin Guan, Peihao Wu, Bryan Dai, Liang Xiao, Bin Dong
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Compare natural-language proof stage vs formal Lean stage | DeepSeek-R1 · FATE-H | accuracy: 71.0% -> 0.0% | Improved |
| Compare natural-language proof stage vs formal Lean stage | DeepSeek-Prover-V2 · FATE-X | accuracy: 9.0% -> 0.0% | Hurt |
| Move from FATE-M to FATE-H to FATE-X | DeepSeek-R1 · FATE-M/H/X | NL accuracy: 94.7% -> 71.0% -> 33.0% | Hurt |
| Use general reasoning model instead of specialized prover | DeepSeek-R1 vs DeepSeek-Prover-V2 · FATE-H | NL pass@1 accuracy: 39.0% -> 71.0% | Improved |
| Remove Lean generation requirement from prompt (ablation) | DeepSeek-R1 · FATE-X | NL accuracy: 33% -> 39% | Improved |
| Remove Lean requirement and formal statement (ablation) | DeepSeek-R1 · FATE-X | NL accuracy: 33% -> 43% | Improved |
| Show formal statement in prompt (ablation) | DeepSeek-R1 · FATE-X | NL accuracy: 43% -> 39% | Hurt |
| Add math-before-Lean output instruction (ablation) | DeepSeek-R1 · FATE-X | NL accuracy: 33% -> 32% | No clear effect |
| Compare Mathlib hallucination rate across models | DeepSeek-R1 vs DeepSeek-Prover-V2 · FATE-H | error rate: 35/39 vs 70/71 | Mixed |
| Compare misalignment errors | DeepSeek-Prover-V2 vs DeepSeek-R1 · FATE-H | misalignment: 3/39 vs 0/71 | Hurt |
| Increase problem difficulty FATE-M to FATE-X | DeepSeek-R1 · FATE | output tokens: 3845.0 -> 10625.1 -> 16597.4 | Mixed |
Closest papers by meaning
- FaithSieve: Fine-Grained Evaluation of Math Proofs with Faithful Formal Evidence (NeurIPS 2026 · similarity 0.88)
- Not All Proofs Are Equal: Evaluating LLM Proof Quality Beyond Correctness (NeurIPS 2026 · similarity 0.87)
- RealMath: A Continuous Benchmark for Evaluating Language Models on Research-Level Mathematics (NeurIPS 2025 · similarity 0.87)
- MathGAP: Out-of-Distribution Evaluation on Problems with Arbitrarily Complex Proofs (ICLR 2025 · similarity 0.87)
- UGMathBench: A Diverse and Dynamic Benchmark for Undergraduate-Level Mathematical Reasoning with Large Language Models (ICLR 2025 · similarity 0.87)
- Llemma: An Open Language Model for Mathematics (ICLR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
