AI Paper Atlas

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

PostTrainBench: Can LLM Agents Automate LLM Post-Training?

Ben Rank, Hardik Bhatnagar, Ameya Prabhu, Shira Eisenberg, Karina Nguyen, Matthias Bethge, Maksym Andriushchenko

ICML 2026 · Large language models › Agents, code and math · approach: Long-horizon LLM agents · 0 citations (OpenAlex)

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
use Opus 4.7 agent instead of base model alonePostTrainBench agents · PostTrainBench weighted averageweighted average score (%): 7.5 -> 27.9Improved
agent post-training versus official instruction tuningbest agent · PostTrainBench weighted averageweighted average score (%): 27.9 vs 51.1Hurt
agent targeted post-training for function callingGemma-3-4B · BFCLaccuracy (%): 67 -> 89Improved
agent targeted post-training on GPQAGemma-3-4B · GPQAaccuracy (%): 31 -> 33Improved
use high instead of medium reasoning effort (ablation)GPT-5.1 Codex Max · PostTrainBench weighted averageweighted average score (%): 19.1 -> 17.4Hurt
use low instead of medium reasoning effort (ablation)GPT-5.1 Codex Max · PostTrainBench weighted averageweighted average score (%): 19.1 -> 15.6Hurt
use high instead of medium reasoning effort (ablation)GPT-5.3 Codex · PostTrainBench weighted averageweighted average score (%): 13.49 -> 17.41Improved
use larger Claude model (Haiku -> Sonnet -> Opus) (ablation)Claude Code agent · PostTrainBench weighted averageweighted average score (%): 6.4 -> 10.7 -> 15.8Improved
increase time budget from 1h to 10h (ablation)agents (Opus 4.5, GPT-5.1 Codex Max, Gemini 3 Pro) · PostTrainBench weighted averageweighted average score (%): about 10-12% at 1h; Opus 4.5 plateaus after 5hImproved
use OpenCode instead of native Codex CLI scaffold (ablation)GPT-5.1 Codex Max · PostTrainBench weighted averageweighted average score (%): 19.1 -> 7.7Hurt
use OpenCode instead of native Gemini CLI scaffold (ablation)Gemini 3 Pro · PostTrainBench weighted averageweighted average score (%): 18.1 -> 14.9Hurt
use OpenCode instead of Claude Code scaffold (ablation)Claude Opus 4.5 · PostTrainBench weighted averageweighted average score (%): 15.8 -> 16.6No 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