Model WarsJuly 5, 2026via The Decoder
Baidu's "Unlimited OCR" processes dozens of document pages in one pass by treating memory like human forgetting
Why it matters
Baidu's architectural breakthrough in attention mechanisms addresses a hard constraint in document AI, signaling a shift in how models manage context and memory trade-offs. This matters for enterprise document processing workflows and sets a new performance bar for the OCR benchmark.
Key signals
- Baidu Unlimited OCR processes 30+ document pages in single pass (vs. previous 10-page limit)
- Modified attention mechanism maintains flat memory usage regardless of page count
- Top score on primary OCR benchmark
- Architecture treats memory like 'human forgetting' — selective retention mechanism
The hook
Baidu just 3x'd OCR capacity. A modified attention mechanism processes 30+ pages where competitors max out at 10—and memory stays flat.
Baidu's Unlimited OCR reads dozens of document pages in a single pass, where previous systems topped out at about ten. A modified attention mechanism keeps memory use flat no matter how many pages the model processes. It currently holds the top spot on the most important OCR benchmark.