AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

Rethinking LLM Reasoning: From Explicit Trajectories to Latent Representations

Cong Jiang, Xiaofeng Zhang, Fangzhi Zhu, XiaoWei Chen, Junxiong Zhu, Zheng Zhang

ICLR 2026 · Large language models › Reasoning and chain of thought · approach: Search and self-correction in reasoning

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Survey-style rethinking of LLM reasoning that contrasts explicit reasoning trajectories with latent representations to reduce inference cost; abstract only available.

Problem

Reasoning trajectories generated by LLMs are often much longer than final answers, causing substantial inference cost even for simple tasks.

Approach

Frames the field as a move from explicit step-by-step reasoning trajectories toward latent reasoning representations; the abstract is truncated so the specific proposed method is not stated.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace explicit reasoning with latent reasoning tokens from a reasoning networkDeepSeek-R1-Distill-Qwen-1.5B, 512-token budget · 5-benchmark averageaccuracy: 37.93 -> 44.91Improved
use LRT instead of LC-R1 at 1024-token budgetDeepSeek-R1-Distill-Qwen-1.5B · 5-benchmark averageaccuracy: 48.53 -> 48.58No clear effect
use LRT instead of baseline at 1024-token budget on AMCDeepSeek-R1-Distill-Qwen-1.5B · AMCaccuracy: 42.88 -> 42.50No clear effect
convert Qwen3-1.7B thinking mode to latent reasoning vs non-thinking, pass@1Qwen3-1.7B · 5-benchmark averageaccuracy: 46.93 -> 48.42Improved
convert Qwen3-4B thinking mode to latent reasoning vs non-thinking, pass@4Qwen3-4B · 5-benchmark averageaccuracy: 65.78 -> 71.60Improved
use LRT instead of Qwen3-8B non-thinkingQwen3-8B · 5-benchmark averageaccuracy: 56.44 -> 59.63Improved
increase latent tokens from 64 to 256 (ablation)LRT, Qwen3-1.7B · 5-benchmark averageaccuracy: 42.53 -> 48.42Improved
increase latent tokens from 256 to 512 (ablation)LRT, Qwen3-1.7B · 5-benchmark averageaccuracy: 48.42 -> 46.92Hurt
increase latent tokens from 256 to 512 (ablation)LRT, Qwen3-8B · 5-benchmark averageaccuracy: 59.63 -> 59.97Improved
add RL stage after SFT (ablation)LRT, Qwen3-1.7B · 5-benchmark averageaccuracy: 41.29 -> 48.42Improved
randomly skip 50% of reasoning tokens in the trajectory (ablation)DeepSeek-R1-Distill-Qwen-7B · MATH-500accuracy: 92.8 -> 90.60No clear effect
use LRT instead of thinking mode at 4096 budgetQwen3-8B · 5-benchmark averageaccuracy: 59.63 -> 69.61Hurt

Limitations

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