Sign InOpen Brain
arXivPaperNeeds Review

Prefix Sliding for efficient test-time scaling

Prefix Sliding drops stale reasoning tokens while retaining instructions and recent work, capping memory during long inference and reportedly making existing models 3× faster without training.

arXiv · Aug 26, 2026
Open Source Open MarkdownOpen JSON
Source Summary

Prefix Sliding retains the instruction-and-tools prefix plus a window of the latest few thousand tokens, discarding older intermediate reasoning. This caps memory independently of reasoning length and made existing models **3× faster without training** while maintaining performance.

Practical Implication

Builders operating long-running agents should distinguish durable context from transient reasoning. Keeping stable instructions and current work may be more efficient than retaining every intermediate token; reinforcement learning with the method enabled traces beyond **100,000 tokens**.

Agent-Ready Context
Prefix Sliding retains the instruction-and-tools prefix plus a window of the latest few thousand tokens, discarding older intermediate reasoning. This caps memory independently of reasoning length and made existing models **3× faster without training** while maintaining performance.

Builders operating long-running agents should distinguish durable context from transient reasoning. Keeping stable instructions and current work may be more efficient than retaining every intermediate token; reinforcement learning with the method enabled traces beyond **100,000 tokens**.

The claims come from a new paper, and the supplied material does not name the evaluated models, workloads, or deployment costs. Its ablations beat summarization and a vanilla sliding window, but broader generality remains open.
Connected Context · Feed7 Judgment

Prefix Sliding reframes long-context reasoning as a retention-policy problem: preserve stable instructions, tools, and recent work while dropping stale intermediate reasoning. Its reported speed and memory gains complement methods that reduce how much reasoning is generated, but the absent model, workload, and cost details prevent treating the 3× result as a general deployment expectation.

Learning When to Think: Adaptive Reasoning for Test-Time Compute AllocationAdaptive Reasoning reduces tokens by choosing how long to think, whereas Prefix Sliding bounds memory during arbitrarily long thinking; the methods address complementary compute and retention costs.GLM 5.3 now available on AI GatewayGLM 5.3’s 1M-token window illustrates the capacity-focused alternative, while Prefix Sliding suggests long-running agents may not need to retain their entire reasoning history even when a large window is available.OctoLong: Mid-Training On Cross-Repository Code Contexts Enhances Long-Context ModelingOctoLong argues that useful long context depends on preserving linked program structure during training; Prefix Sliding adds an inference-time question of which durable structure must survive once transient reasoning is discarded.
Context Map
modelcoding#reasoning#context-caching
Uncertainty
The claims come from a new paper, and the supplied material does not name the evaluated models, workloads, or deployment costs. Its ablations beat summarization and a vanilla sliding window, but broader generality remains open.