At its DevDay 2026 in San Francisco, #OpenAI unveiled 25 significant updates across its core product lines, including ChatGPT, Codex, GPT-6 models, and APIs. The most prominent announcements were the introduction of the new always-on agent, Dots, and the significantly enhanced, cost-effective GPT-6.1 Sol model.
Dots: OpenAI’s Always-On Agent
Dots represents OpenAI’s latest venture into always-on agents, designed to function as a persistent digital counterpart with continuous memory. Unlike conventional ChatGPT interactions, Dots can operate continuously. Users define a long-term goal, and Dots maintains context, remembers ongoing priorities, and handles tasks requiring days, weeks, or even longer to complete. It tracks progress, monitors changes, and follows through until completion, fundamentally differentiating it from single-task agents.
Key capabilities of Dots include proactive reporting on progress or changes without needing prompts; continuous tracking of long-term objectives across multiple sessions; and the ability to invoke browsers and applications, operate software interfaces, retrieve information, and execute tasks. Furthermore, it can manage several projects simultaneously.
OpenAI demonstrated various use cases. For instance, in app development, Dots can continuously monitor user feedback, automatically organize requirements, fix bugs, run tests, and ultimately present complete PRs and change videos to developers. For product launches, it can understand product positioning and team requirements, adapting release materials and documentation as scope changes. In research, when new data arrives, Dots can rerun analyses, investigate anomalies, update paper charts, and modify related explanations. A real-world example highlighted an early test user who forgot to invoice a publisher; their Dot detected this, prepared the invoice, and sent it after human confirmation.
Dots can integrate with ChatGPT, Slack, and Microsoft Teams, proactively reporting progress, asking questions, or flagging decisions for human intervention. For safety, Dots defaults to read-only tool access, preventing it from sending messages or modifying applications independently. For external operations, the system acts based on user settings—automatic execution, approval requests, or prohibition. Sensitive actions like password changes always require user confirmation. Dots is currently available to Pro and Business Premium users, with plans to expand to multiple collaborative Dots in the future.
GPT-6.1 Sol: Astra-level Capability at a Fraction of the Cost
Recognizing the critical role of per-unit cost for always-on agents, OpenAI introduced GPT-6.1 Sol, a significant upgrade to GPT-6 Sol. Officially, it approaches the intelligence level of GPT-6 Astra, yet its standard input and output Token fees are only one-fifth of Astra's.
In the DeepSWE v1.1 benchmark, which evaluates long-chain software engineering tasks on real codebases, GPT-6.1 Sol matched GPT-6 Astra's performance, improving upon GPT-6 Sol by up to 6.4 percentage points. In the OSWorld 2.0 Computer Use test, it showed a 7 percentage point improvement over GPT-6 Sol, trailing Astra by just 2.1 percentage points. Its standard API pricing is $2 per million input tokens, $0.1 for cached input, and $10 for output tokens.
OpenAI specifically highlighted GPT-6.1 Sol's enhanced capabilities in Agentic Coding, Computer Use, and Professional Work—precisely the types of tasks that consume the most tokens for persistent agents. By releasing a model with near-flagship performance at a substantially reduced cost, OpenAI aims to complement Dots' continuous operation model, dramatically cutting operational expenses and forming a sophisticated combined product strategy.
Other notable updates include a major upgrade to Codex, evolving towards a "cloud-based software engineering team"; the Agents API gaining Computer Use and Codex capabilities; the launch of ChatGPT Space for team collaboration; significant improvements to Plugins, further opening the ChatGPT platform; and Sign in with ChatGPT, enabling users to log into external products like Devin, Notion, and OpenClaw with their ChatGPT accounts.
[AgentUpdate Depth Analysis]OpenAI's DevDay 2026 clearly signals a strategic shift towards "always-on, cost-effective, and highly efficient" AI agents. The introduction of Dots fundamentally alters the landscape by moving beyond single-shot LLM interactions to stateful, persistent task management across sessions. This addresses a critical limitation in existing agent frameworks like LangChain or CrewAI regarding memory and continuity. Coupled with GPT-6.1 Sol's Astra-level performance at one-fifth the cost, OpenAI tackles the economic barrier to widespread agent deployment. High token consumption previously restricted complex real-world applications. Now, the synergy between low-cost models and persistent agents empowers businesses and individuals to deploy intelligent systems for long-duration, high-frequency tasks at an affordable rate. This will accelerate the transition of AI agents from experimental labs to practical production environments. We anticipate a surge in Agentic Coding and Computer Use applications, transforming agents from mere information tools into autonomous, proactive, and continuously learning digital collaborators, revolutionizing software development, customer service, and personal productivity. This move will also spur further maturation and standardization of the entire AI agent toolchain, fostering innovative applications.
