Loading…
The item combines a leaderboard revision with model updates, but the cited OCR scores remain vendor-reported and the roundup provides limited independent validation.
Limits of the evidence: Discovery-class roundup; Baidu's benchmark scores are self-reported, and the source says a public MarathonDocBench release was not found.
What to watch
Assessment revised Oct 29, 2026
How to read the top model leaderboard: Who pays the most for AI inference and context?
Andrew NgOct 29, 2026First report
First assessment, from Andrew Ng
No maturity ladder applies to this desk.
Sens.ai aggregates and assesses published reporting. The assessment above is machine generated; the original sources are authoritative.