AI Paper Atlas

Research map / Large language models / Agents, code and math

Agnostics: Learning to Synthesize Code in Any Programming Language with a Universal Reinforcement Learning Environment

Aleksander Boruch-Gruszecki, Yangtian Zi, Zixuan Wu, Tejas Oberoi, Carolyn Jane Anderson, Joydeep Biswas, Arjun Guha

ICLR 2026 · Large language models › Agents, code and math · approach: Software engineering agents

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 changedWhereMeasuredEffect
train with Agnostics RL on Ag-Codeforces-X (Fortran)Qwen3-4B · Ag-LiveCodeBench-Fortranpass@1: 0 -> 15Improved
train with Agnostics RL on Ag-Codeforces-X (Lua)Qwen3-4B · Ag-LiveCodeBench-Luapass@1: 11 -> 23Improved
scale base model from 4B to 8BAgnostics-trained Qwen3 · Ag-LiveCodeBench-Rpass@1: 15 -> 19Improved
train on easier MBPP problems instead of CodeforcesQwen3-4B · Ag-LiveCodeBench-Rpass@1: 10 -> 9Hurt
train on easier MBPP problems instead of CodeforcesQwen3-4B · MultiPL-E Juliapass@1: 51 -> 62Improved
train on Codeforces rather than MBPPQwen3-4B · Ag-LiveCodeBench-Luapass@1: 15 -> 23Improved
apply Agnostics to non-Qwen modelDeepSeek Coder 6.7B Ins · MultiPL-E Rpass@1: 37 -> 52Improved
apply Agnostics to non-Qwen modelSmolLM3 3B · Ag-LiveCodeBench-Luapass@1: 1 -> 8Improved
RL training via Agnostics instead of distilling from Sonnet 4 ThinkingQwen3-4B Fortran · Ag-LiveCodeBench-Fortranpass@1: 2 -> 15Improved
add partial reward for runnable but wrong programs (ablation)GRPO reward · Ag-LiveCodeBench-Luapass@1: 23.00 -> 18.57Hurt
add partial reward for runnable but wrong programs (ablation)GRPO reward · Ag-Codeforces-Lua testpass@1: 24.76 -> 13.33Hurt
increase GRPO group size from 32 to 64 (ablation)Qwen3-4B-CF-Lua · Ag-LiveCodeBench-Luapass@1: 23.00 vs 21.71No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy