LlamaIndex open-sources ParseBench-2 leaderboard ranking 20+ open-weight OCR models
llama_index · x · 2026-11-18
LlamaIndex's Jerry Liu announced ParseBench-2, a public document-parsing/OCR benchmark that has evaluated 20+ open-weight models, with the leaderboard hosted on Hugging Face and 30k monthly downloads. Using human-verified pages and deterministic rules, it scores models across tables, charts, content faithfulness, semantic formatting, and visual grounding. KDLAI/KDL-Frontier-Parser-nano-v2 leads at 78.89, ahead of Infinity-Parser2-Pro (77.08) and LightOnOCR-3-4B (77.02); Qwen3-B-27B ranks 9th. The team is iterating on the benchmark and adding models rapidly.
More from Models
- Noam Brown's first take on Astra 6.2: expect math breakthroughs within weeks — polynoamial · 2026-11-18
- Google set to release another Gemini model, targeting Anthropic over OpenAI — kimmonismus · 2026-11-18
- Google building new robotics model Astra with 1M-token context, config leaks — lyraxana · 2026-11-18
- Leak: Google to unveil 'krypton' tomorrow, claimed to beat Fable 5.2 and GPT Astra 6.2 at lower cost — imjustnewatai · 2026-11-18
- GPT-6.2 Astra tops robot-code test: 2.5B net solving 79% of LIBERO, Python code hits 96.8% — burny_tech · 2026-11-18
- Goodfire's Neuron Communities: tiny neuron groups in LLMs drive behaviors like arithmetic — burny_tech · 2026-11-18