Skip to content
VECTOR WIREAI INTELLIGENCE
⌘K
UTC
Models795Deals30d241Policy30d72, 16 pendingSources1,108

AI Research Radar

Monday, May 4: 1 frontier paper out of 175 read.

1 frontier18 cleared175 papers read

Frontier

Papers rated as moving the field

How we choose

Jailbroken Frontier Models Retain Their Capabilities

“Evaluating 28 jailbreaks on five benchmarks across Claude models ranging in capability from Haiku 4.5 to Opus 4.6, we find Haiku 4.5 loses an average of 33.1% on benchmark performance when jailbroken, while Opus 4.6 at max thinking effort loses only 7.7%.”

The authors, in the abstract

AnthropicarXivPDF

This week in research

Apr 28 – May 4: 127 cleared, 12 frontier

The week’s papers have not been sorted into fields yet.

Models under test

Papers this week that name the model

  1. GPT-4o5
  2. GPT-5.43
  3. GPT-53
  4. Claude Opus 4.62
  5. GPT-5.3-Codex1
  6. Gemini 3 Pro1
  7. Claude Sonnet 4.61
  8. Claude 3.5 Haiku1

Labs publishing

Papers this week with an author at the lab

  1. AllenAI3
  2. NVIDIA3
  3. Alibaba1
  4. Anthropic1
  5. Google1
  6. Meta1
  7. Microsoft1
  8. Mistral1

Also cleared

17 papers

All 17 papers
SIGNIFICANT ON THE BOARDS
MODEL TRACKER2,648 new benchmark results across 174 models this weekDEAL FLOW$158B disclosed AI funding this week across 61 roundsPOLICY125 policy events tracked across 7 days