SFR-DR-20B achieves 28.7% on Humanity's Last Exam full text-only benchmark, 65% relative improvement over gpt-oss-20b baseline.

{
  "doc": "nguyen-2025-sfr-deepresearch.md",
  "context": "Main evaluation result showing best variant outperforms many proprietary and open-source baselines of comparable or larger sizes.",
  "norm_label": "SFR-DR-20B achieves 28.7% on Humanity's Last Exam full text-only benchmark, 65% relative improvement over gpt-oss-20b baseline.",
  "graphify_id": "finding_sfr_dr_20b_performance",
  "source_file": "nguyen-2025-sfr-deepresearch.md",
  "imported_from": "/Users/antonborzov/Documents/Research.nosync/papers/extract_typed_out/nguyen-2025-sfr-deepresearch/graph.json",
  "extracted_type": "finding",
  "source_location": "§4.1, Table 1",
  "graphify_file_type": "finding"
}