My notis

The register Topic

Benchmarks

7 stories · first 26 Sep · latest 8 Nov

Ongoing case: Agents that cheat or mislead Ongoing case: Anthropic's Claude models Ongoing case: Google's Gemini models Ongoing case: OpenAI's models Ongoing case: Typed decision models and APIs Compare with another company or topic

All 7 stories

8 Nov ResearchOpenAI publishes LifeVal, a personal-assistant test drawn from time-use surveys
8 Nov ResearchHow METR's time horizon rates AI agents by human task length
2 Nov ResearchArtificial Analysis's leaderboard ranks Gemini 4 Carbon (high) first with an Intelligence Index score of 59
13 Oct Models and releasesMeta releases Muse Wonder and Muse Wonder Full for developers and subscribers
5 Oct ResearchCheatBench authors say every tested agent cheats in some settings
28 Sep ResearchProgramDistill, a benchmark built from 26 web apps, scores GPT-6 Astra at 49.2 percent and Claude Opus 5 at 28.8.
26 Sep ResearchPaper finds prompt injection can shift Jev's typed decisions, though rarely to the attacker's target

Stories per week

121 Sep to 27 Sep

128 Sep to 4 Oct

15 Oct to 11 Oct

112 Oct to 18 Oct

019 Oct to 25 Oct

026 Oct to 1 Nov

32 Nov to 8 Nov

09 Nov to today

Counts cover only the stories we published, not all AI news. Small numbers can mislead.

Named together with