OpenAI DevDay 2026: Sam Altman Unveils Always-On 'Dots' Agents, GPT-6.1 Sol, and Ultrafast Tier

During the opening keynote of OpenAI DevDay 2026 at Fort Mason in San Francisco, CEO Sam Altman announced more than 20 major platform updates headlined by 'Dots'—always-on autonomous personal AI agents that learn user preferences and handle ongoing real-world responsibilities in the background. OpenAI also introduced GPT-6.1 Sol delivering near-Astra frontier intelligence at one-fifth the token cost, an Ultrafast inference tier generating 300 tokens per second, and native computer use integrated directly into the Agents API.

Dots: Always-On Autonomous Agents and GPT-6.1 Sol

Headlining the conference was the debut of Dots, OpenAI shift from passive chatbots to active, persistent personal agents. Operating continuously across personal and enterprise workflows, a Dot learns user priorities, monitors incoming events, and independently coordinates multi-step tasks across web tools without requiring prompt-by-prompt guidance. Powering these agentic interactions is the new GPT-6.1 Sol foundation model, which delivers significant performance uplifts in automated code generation, desktop computer interaction, and professional reasoning at 80% lower token pricing than GPT-6 Astra.

Key Keynote Announcements

• Dots Autonomous Agents: Always-on background agents handling ongoing workflows for Pro, Business, and Enterprise users.
• GPT-6.1 Sol Architecture: Upgraded reasoning model delivering frontier intelligence at 1/5th of standard input and output token costs.
• Ultrafast Speed Tier: Premium inference speed delivering up to 300 tokens per second (8x acceleration in Codex, 6x in API).
• Agents API with Computer Use: Native developer endpoints supporting direct GUI interaction, multi-agent coordination, and context compaction.
• ChatGPT Space & Pages: Collaborative team workspace where humans, ChatGPT, and autonomous Dots edit shared interactive documents.

300 Tokens/Sec Ultrafast Tier and Native Computer Use APIs

To meet developer demand for instantaneous code generation and interactive agents, OpenAI launched Ultrafast, a dedicated high-throughput inference tier capable of generating up to 300 tokens per second in Codex. Developers also gained access to the expanded Agents API with native computer use, allowing models to perceive screen pixels, click interface elements, and fill forms within virtualized sandboxes. Through a new partnership with Amazon Web Services, these agentic capabilities will also deploy natively via Bedrock Managed Agents.

Event Profile Sheet