Databricks Launches Precision Mode, Claiming 94.7% Accuracy in AI Document Extraction
Summary
Databricks launches Precision Mode for its AI Extract API, achieving 94.7% accuracy across 9,000 complex documents and outperforming GPT-4.5 Sol by seven percentage points using fine-tuned models and parallel agentic processing.
Key Points
- Databricks is launching Precision Mode in its AI Extract document extraction API, setting a new accuracy benchmark for complex enterprise document processing across long documents, large nested outputs, and reasoning-heavy schemas.
- Precision Mode combines custom fine-tuned extraction models with an agentic harness that decomposes large extraction jobs, runs subagents in parallel, and merges results into a single structured output, overcoming common failure modes like chunk timeouts and truncated outputs.
- Across roughly 9,000 complex documents spanning 15 benchmarks, Precision Mode achieves 94.7% extraction accuracy, outperforming the next best frontier model, GPT-5.6 Sol, by seven percentage points.