OpenAI’s dots score 8.8 in hands-on test, rank third among five agents
OpenAI’s new AI agents, called dots, scored an average of 8.8 out of 10 in an independent October 2026 test across ten business tasks. The agents, built on GPT-6 Astra, run persistently in the cloud with dedicated Linux environments and browser access. They connect to over 4,000 applications via plugins and are designed for multi-step tasks, learning user preferences over time. Access began for…
Key points
- OpenAI’s dots scored 8.8/10 in an independent test, ranking third behind Claude Cowork (9.3)
- Dots run persistently in cloud with Linux/browser access, linking to 4,000+ apps via plugins
- Internal safety data: blocked 16,600 simulated email attacks but had 19.7% scope violations in 10-task chains
The review found dots outperformed in user experience (9.2) but trailed Claude Cowork (9.3) in overall performance. OpenAI’s internal safety data shows dots blocked all 16,600 simulated email attacks but had higher scope violations—8.6% for five-task chains and 19.7% for ten-task chains—indicating occasional task drift.
Model page: GPT-6 Astra →
OpenAI’s Dots score well in hands-on test, but rank third among five agents
cryptobriefing.com · 26 October 2026
Loading the full article…
This text was published by cryptobriefing.com and written by Diego Almada Lopez. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Wired tests six AI agents, ranks Gemini Spark top for news briefings · 1 src
- Developer builds voice-driven AI agent with ledger, multiple PCs and Claude Code · 1 src
- IBM adds governed identity and credential controls for enterprise AI agents · 1 src
- LiteLLM 1.107.0 compromised, developer says release process was hijacked · 1 src
- Claude Code says 'Done' but it doesn't work, lecture explains · 1 src
Comments
via GitHub Discussions