PulseAugur
EN
LIVE 11:56:10
中文(ZH) KV Cache被压成常数！OCR开源新王：页数没上限，缓存砍96%

Baidu's Unlimited OCR 2 cuts cache use by 96% for long documents

By PulseAugur Editorial · [1 sources] ·

Baidu has released and open-sourced Unlimited OCR 2, an upgraded version of its optical character recognition model. This new version significantly reduces KV cache usage by approximately 96%, bringing it down to 1.77GB from 47GB for hundred-page documents. This optimization is achieved through a sparse reference sliding window attention mechanism (SR-SWA) and an unlimited decoding paradigm, allowing the model to process extremely long documents efficiently without a proportional increase in memory or computation. AI

IMPACT This release significantly improves long-document processing efficiency, potentially accelerating adoption in fields like legal tech, publishing, and archival.

RANK_REASON New model release from a major AI lab (Baidu) with significant technical advancements. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on 量子位 (QbitAI) →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

Baidu's Unlimited OCR 2 cuts cache use by 96% for long documents

How we ranked this

Signal score
33 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Significant
New model release from a major AI lab (Baidu) with significant technical advancements. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
model release, product
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [1]

  1. 量子位 (QbitAI) TIER_1 中文(ZH) · 听雨 ·

    KV Cache Reduced to a Constant! OCR's New Open-Source King: Unlimited Pages, 96% Cache Reduction

    OCR 3.0时代，来了

RELATED ENTITIES

RELATED TOPICS