LlamaIndex Launches Extract v2.5, Beats Claude and GPT at 30%-4x Lower Cost
llama_index · x · 2026-10-02
LlamaIndex CEO Jerry Liu announced Extract v2.5, a set of frontier document-extraction agents in three tiers (cost-effective, agentic, agentic plus), tuned for value accuracy and grounding. Key gains on the agentic tier: long lists 86.1%→95.5%, records spanning pages 85.5%→96.5%, scanned forms 90.9%→95.7%. The agents reportedly outperform Opus 5.5 and GPT-6 Sol while costing 30%-4x less, and are SOTA on price-performance across ExtractBench. New features include advanced citations (bounding boxes for supporting values) and structural reasoning that adapts extraction algorithms to document type and layout.
More from coding & agent
- Surrendering to Claude Code's permission prompts: granting Bash(*) to all — Daniel_Farinax · 2026-10-02
- Prime Intellect ships full post-training stack as Extropic runs custom RL with it — beffjezos · 2026-10-02
- Build an agent that remembers with Agent Platform Memory Bank — rseroter · 2026-10-02
- 8 Open-Source AI Agent Tools: From Computer-Use to Web Scraping — Aiden_Tech_Ai · 2026-10-02
- Claude Code 2.1.287 is about to be released — ClaudeCodeLog · 2026-10-02
- Claude Code creator uses "opponent" sub-agents to cross-check each other's work — every · 2026-10-02