AI Paper Atlas

Research map / Video / Video-language understanding

SAGE: Training Smart Any-Horizon Agents for Long Video Reasoning with Reinforcement Learning

Jitesh Jain, Allen AI blank, Jialuo Li, Zixian Ma, Jieyu Zhang, Chris Dongjoo Kim, Sangho Lee, Rohun Tripathi, Tanmay Gupta, Christopher Clark, Humphrey Shi

CVPR 2026 · Video › Video-language understanding · approach: Event and anomaly reasoning in video · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add RL after SFT (ablation)SAGE (Qwen3-VL-4B) · SAGE-Benchoverall accuracy (%): 64.6 -> 68.4Improved
add RL after SFT (ablation)SAGE (Qwen3-VL-8B) · SAGE-Benchoverall accuracy (%): 63.9 -> 68.0Improved
add the ground-event tool (ablation)SAGE (Qwen3-VL-8B) · SAGE-Benchoverall accuracy (%): 67.3 -> 68.0Improved
add the web-search/parse-website tool (ablation)SAGE (Qwen3-VL-8B) · SAGE-Benchoverall accuracy (%): 65.5 -> 68.0Improved
add the analyze tool (ablation)SAGE (Qwen3-VL-8B) · SAGE-Benchoverall accuracy (%): 63.4 -> 68.0Improved
add the extract-video-parts tool (ablation)SAGE (Qwen3-VL-8B) · SAGE-Benchoverall accuracy (%): 63.0 -> 68.0Improved
add the transcribe-speech tool (ablation)SAGE (Qwen3-VL-8B) · SAGE-Benchoverall accuracy (%): 62.5 -> 68.0Improved
give access to the video (ablation)SAGE (Qwen3-VL-8B) · SAGE-Benchoverall accuracy (%): 41.0 -> 68.0Improved
use agentic mode instead of direct mode with SFT+RL (ablation)SAGE (Qwen3-VL-4B) · SAGE-Benchoverall accuracy (%): 66.3 -> 68.4Improved
wrap Gemini-2.5-Flash in the SAGE-MM systemGemini-2.5-Flash · SAGE-Benchoverall accuracy (%): 68.1 -> 71.3Improved
wrap GPT-4o in the SAGE-MM systemGPT-4o · SAGE-Benchoverall accuracy (%): 71.6 -> 73.4Improved
SFT only on Qwen3-VL-8B vs base modelSAGE (Qwen3-VL-8B) · SAGE-Benchoverall accuracy (%): 64.9 -> 63.9Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy