OpenAI's Dots takes the lead on PersonalAgentBench as task set doubles
adcock_brett · x · 2026-10-13
- @harklabs and OpenAI's Dots join PersonalAgentBench, with Dots taking #1, Hark close behind, and Gemini Spark sliding to third, followed by Instinct, Grok Bot, and Muse.
- The task set doubled this week, adding research, image understanding, editing, and everyday communication. Gemini Spark was strongest on research but weaker on image/editing; Hark and Gemini tied on user-authority tasks; Dots was most consistent overall though weaker at inferring personal context. No assistant led every category.
- Brett Adcock quote-posted it saying it's embarrassing and he'll work to crack #1.
More from coding & agent
- Whodunit: an AI-directed murder mystery you play by voice on a video call — threepointone · 2026-10-13
- Musk amplifies user claim that Grok Bot autonomously canceled 7 SaaS subscriptions — elonmusk · 2026-10-13
- Patchbay: a StackOverflow-for-agents where stuck WebMCP agents ask for help — seanwbren · 2026-10-13
- 30% of my Codex chat time goes to context compaction, researcher complains — scaling01 · 2026-10-13
- 67-minute hands-on AI agent tutorial: from workflows to competitor tracking — goyalshaliniuk · 2026-10-13
- Rivet-KAG: open-source hybrid vector + graph RAG with verifiable citations — No-Application-1232 · 2026-10-13