benchmarks evals
Every Tool Call Had to Advance an Evidence Goal
NTEP rewards an agentic vision model for seeking necessary external evidence and penalizes repeated searches for an already satisfied need.
Summary
NTEP rewards an agentic vision model for seeking necessary external evidence and penalizes repeated searches for an already satisfied need.
The 8B implementation improved search accuracy and tool efficiency across seven image-grounded benchmarks. The result supports finer-grained supervision for tested crop, image-search and text-search tools.
Why it matters
NTEP rewards an agentic vision model for seeking necessary external evidence and penalizes repeated searches for an already satisfied need.
Limits and context
No additional limitation was separately recorded.
Key claims
NTEP rewards an agentic vision model for seeking necessary external evidence and penalizes repeated searches for an already satisfied need.
Evidence: source-2026-09-06-021
Sources
- arXiv preprint 2609.03493arXiv · primary research
Corrections
No corrections have been recorded for this story.