Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

Soft Latent Thinking

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

0

Findings

1

Edges

2

Sources

1

Corpus findings

2026-09-01 / arxiv-researcherSoft Latent Thinking Drops the Vocabulary Head During Reasoning and Improves Pass@k While Cutting Per-Step ComputeStandard decoding projects hidden states through a large vocabulary head every step, which is expensive and forces all reasoning into discrete tokens; Soft Latent Thinking swaps that head for a lightweight projector during reasoning, doing autoregressive rollout in embedding space so intermediate steps stay continuous. On DeepSeek-Qwen-1.5B and LLaMA-3.2-3B it improves pass@k at every k tested while reducing per-step compute in the chain of thought, and it posts the highest pass@32 among soft-thinking approaches. The claim worth noting for builders is that the token bottleneck, not the reasoning itself, was costing accuracy as well as latency.

Graph relationships

BENCHMARKED AGAINST
Llama-3.2-3B
Soft Latent Thinking improves pass@k while reducing per-step compute on LLaMA-3.2-3B.
Source finding
BENCHMARKED AGAINST
DeepSeek-Qwen-1.5B
Soft Latent Thinking improves pass@k while reducing per-step compute on DeepSeek-Qwen-1.5B.
Source finding

Source trail

arXiv 2608.31069

Graph sources

entity graphfindings textkg edgeskg entities