AI Paper Atlas

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

MindZero: Learning Online Mental Reasoning With Zero Annotations

Shunchi Zhang, Jin Lu, Chuanyang Jin, Yichao Zhou, Zhining Zhang, Tianmin Shu

ICML 2026 · Large language models › Agents, code and math · approach: Reasoning benchmarks and platforms · 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
add explicit LLM prior modeling to the reward (ablation)MindZero Qwen3-4B · Household proactive assistancespeedup (%): 17.0 -> 19.1Improved
maintain multiple hypotheses instead of a single most likely one (ablation)MindZero Qwen3-4B · Household proactive assistancespeedup (%): 10.3 -> 19.1Improved
add entropy bonus to the reward (ablation)MindZero Qwen3-4B · Household proactive assistancespeedup (%): 5.2 -> 19.1Improved
add multiple hypotheses (compute cost) (ablation)MindZero Qwen3-4B · Household proactive assistanceTFLOPs: 132.6 -> 201.2Hurt
train with MindZero self-supervised RLQwen3-VL-4B · GridWorld QAaccuracy: 37.7 -> 95.0Improved
train with MindZero self-supervised RLQwen3-VL-8B · GridWorld proactive assistancespeedup (%): -0.1 -> 24.5Improved
train with MindZero self-supervised RLQwen3-4B · Household proactive assistancespeedup (%): 2.3 -> 19.1Improved
train Llama-3.2-3B with MindZero after format warm-upLlama-3.2-3B · Household proactive assistancespeedup (%): 2.3 -> 4.3No clear effect
train Qwen3-4B with MindZero (human study)Qwen3-4B · Household human experimentspeedup (%): 2.62 -> 19.70Improved
use Gemini-3-Flash instead of MindZero Qwen3-4Bhelper agent · Household human experimentspeedup (%): 19.70 -> 23.40 (p=0.24)No clear effect
use thinking mode for GPT-5.2GPT-5.2 · Household QAaccuracy: 65.0 -> 73.5Improved
use thinking mode for Gemini/GPT-5.2 on GridWorldGPT-5.2 · GridWorld QAaccuracy: 50.7 -> 50.7No 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