Skip to content
VECTOR WIREAI INTELLIGENCE
UTC

AI Research Radar

Friday, November 20: 5 frontier papers out of 449 read.

5 frontier83 cleared449 papers read

Frontier

Papers rated as moving the field

How we choose

MKA

On the MKA benchmark, Claude Fable 5.2 achieves 56.6% Climb-WSR, with Gemini 4 Carbon close behind.

Evaluating Multimodal Knowledge Agents on end-to-end professional tasks

AgentsGooglearXivPDF

PhyDEX

Three vision-language-action models evaluated on PhyDEX all achieve success rates below 3%, with ground-truth physical properties roughly doubling success.

RoboticsBenchmark

Keel

Keel's weight edits remove most loops on Qwen3.5-0.8B and Qwen3.5-2B across six image benchmarks and raise average accuracy.

Vision

Beam Model Technical Report

Beam refuses or pivots on 88.9% of harmful requests while complying with 81.5% of harmless requests, averaged across conversation turns and harm categories.

Training

This week in research

Nov 14–20: 339 cleared, 18 frontier

Also cleared

78 papers, by field

Security & privacy24
All 24 papers in Security & privacy
AI & society11
All 11 papers in AI & society
Safety & alignment9
All 9 papers in Safety & alignment
Agents8
All 8 papers in Agents
Robotics & embodied AI8
All 8 papers in Robotics & embodied AI
Inference & efficiency6
Vision & multimodal4
Language & speech3
Training & data2
Reasoning & math1
Science & health1
Machine learning1