Research map / Large language models / Agents, code and math
ALGO: Synthesizing Algorithmic Programs with Generated Oracle Verifiers
Kexun Zhang, Danqing Wang, Jingtao Xia, William Yang Wang, Lei Li
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 |
|---|---|---|---|
| rerank Codex programs with ALGO verifier instead of CodeT | Codex · CodeContests | 1@k: 2.10 -> 5.60 | Improved |
| rerank Codex programs with ALGO verifier vs raw Codex | Codex · CodeContests | 1@k: 0.70 -> 5.60 | Improved |
| add ALGO verifier to PG-TD (replace generated tests) | PG-TD · CodeContests | 1@k: 0.67 -> 1.57 | Improved |
| add ALGO oracle verifier to ChatGPT | ChatGPT · CodeContests | 1@k: 4.40 -> 12.00 | Improved |
| increase n toward k (100@k) | ALGO w/ Codex · CodeContests | n@k gain over Codex: +4.90 -> +3.60 | Mixed |
| replace example tests with ALGO generated tests for ChatGPT coder (ablation) | ChatGPT coder · LeetCode | 1@20: 4.4% -> 12.0% | Improved |
| replace ChatGPT-generated tests with ALGO tests (ablation) | ChatGPT coder · LeetCode | 1@20: 6.8% -> 12.0% | Improved |
| use ChatGPT-generated tests instead of example tests at 7@20 (ablation) | ChatGPT coder · LeetCode | 7@20: 12.2% -> 11.7% | No clear effect |
| use ALGO generated test cases instead of example cases | verifier · CodeContests | agreement with judge: 57.39% -> 94.78% | Improved |
| use ALGO generated test cases instead of example cases | verifier · LeetCode | agreement with judge: 68.57% -> 74.29% | Improved |
| use ALGO test cases for coverage | verifier · CodeContests | statement coverage: 85.78% -> 89.39% | Improved |
| increase number of generated test cases beyond 30 (ablation) | ALGO w/ Codex · CodeContests | n@k | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CodeT — CodeT: Code Generation with Generated Tests (ICLR 2023) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- AlphaVerus: Bootstrapping Formally Verified Code Generation through Self-Improving Translation and Treefinement (ICML 2025 · similarity 0.88)
- How efficient is LLM-generated code? A rigorous & high-standard benchmark (ICLR 2025 · similarity 0.88)
- Rethinking Verification for LLM Code Generation: From Generation to Testing (NeurIPS 2025 · similarity 0.86)
- ORACLE: Optimizing Reasoning Abilities of Large Language Models via Constraint-Led Synthetic Data Elicitation (AAAI 2026 · similarity 0.86)
- BaxBench: Can LLMs Generate Correct and Secure Backends? (ICML 2025 · similarity 0.85)
- CodeChemist: Test-Time Scaling for Low-Resource Code Generation via Functional Knowledge Transfer (ICML 2026 · similarity 0.85)
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.
