Iβd switch coding agents for a full week of uninterrupted work.
For Claude Code vs Codex, I care about finished tasks before the cap; including retries and fixes. API-equivalent dollar totals donβt tell me that.
@Nils_Reimers@cohere For tables, would you expand the selected spans back to the row + headers + linked footnotes? A number can survive pruning while its unit or exception disappears. That's the case I'd want to test.
@VikParuchuri With 2.5 reading table HTML and Surya 2 supplying row/column boxes, how would you align the two for cell-level citations? Repeated values make text matching ambiguous, especially if the models disagree on a merged cell.
@jerryjliu0 , tried Max Effort on dense drawings. Text extraction was solid, but in one test only 20% cell citation boxes pointed to the correct location. We need reliable links back to the drawing. Any tips for improving this?
Today we're introducing Extract Max Effort - the best frontier agent in the world for document extraction π«.
π It is an absolute bazooka. Use this when you absolutely need perfect 100% extraction accuracy over the most complex documents and schemas. This includes reasoning over dense well log chats, full insurance rate filings spanning 100s of pages, or dense construction blueprints.
π It sets a new state of the art on ExtractBench Pro, achieving 61% perfect extraction accuracy. It beats Codex+Astra and Claude Code+Opus 5.5 on price/performance.
It achieves this accuracy by allocating dynamic tokens depending on page and schema complexity. It is a toggle on top of our Agentic Plus mode, so additional pricing depends on any additional compute required.
Blog: https://t.co/fd0eaOkp4N
Come check it out in LlamaParse! https://t.co/XYZmx5TFz8
Benchmark: https://t.co/FkPrNb5iti
Give pipeline agents a cheap way to be wrong.
Databricks' benchmark: ~2.3x the tokens, ~1/34 the Spark executor time with validation before execution.
Measure cost per correct result, retries included.
https://t.co/Lox7NJWjKP