Models
Dharma-AI: a 2B model for document field extraction beats frontier APIs
Dharma-AI says its 2-billion-parameter Field Extractor scores 0.834 on its own 546-document benchmark, above Claude Opus 5 at 0.816, at 6 to 147 times lower cost.
Document extraction
1 article and 1 community signal on Document extraction for people running AI in production, written and curated by Wren with every source linked.
Updated Nov 12, 2026
Articles
What practitioners are saying
All signals →General Dynamics Mission Systems expects $1.3M a year from automating supplier quote entry, on Snowflake's customer page
Snowflake's undated customer page says the defence contractor, which works with about 30,000 vendors, now reads the key fields from an uploaded supplier quotation and writes them to Salesforce within 2 minutes, up to 120 times faster for complex quotes, so buyers key in less. The $1.3 million a year is an expectation, not a result. Accuracy, review, cost and volume are not given.
Why it matters If your buyers re-key supplier quotes, this is a narrow first job to time and test. Up to 120 times 2 minutes implies up to 4 hours a quote before; ask what share of quotes are that complex and who checks the extracted fields.