Prefix Sliding: discarding stale reasoning tokens makes test-time scaling 3x faster
Bedrovelsen · x · 2026-08-27
A new paper proposes Prefix Sliding: most intermediate reasoning tokens lose importance as the model continues reasoning, so retaining the full history may not be worth the cost. The method keeps only the prefix plus a sliding window of the last few thousand tokens.
Results: without training, existing models run 3x faster with maintained performance; training with Prefix Sliding via RL enables scaling to reasoning traces beyond 100k tokens for even better results. Code is open-sourced.
Related event: Prefix Sliding triples inference speed with no quality loss(4 posts)→
More from Infra
- SGLang ships Day-0 support for FLUX 3 Image Klein: 10 reference images, BF16/FP8/NVFP4 variants — ying11231 · 2026-10-16
- NVIDIA's Nemotron 3.5 Super VL hits CoreWeave: 12B-active multimodal MoE at $0.25/M input tokens — wandb · 2026-10-16
- Chip equipment beats bull case: 2026 WFE seen at $153B, $278B by 2030 — BenBajarin · 2026-10-16
- Steam now launches in 5.23s, beating Epic's 5.47s — and vLLM gets roasted for slow startup — AlpinDale · 2026-10-16
- Shadeform to Announce Six New Features on Oct 19: Lower GPU Costs, More Suppliers — _jaydeepkarale · 2026-10-16
- Fastino launches Training API for SFT/RL, claiming 25-85% cost cut vs Tinker — kalyan_kpl · 2026-10-16