Cheaper agents, persistent work, AI rules
Frontier AI just got cheaper, more persistent, and more structured. OpenAI’s DevDay brought three distinct changes for builders, while Google joined a new White House-led self-governance effort with important unanswered questions.
In today’s edition
- GPT-6.1 Sol cuts costs for sustained agents.
- ChatGPT Dots keep cloud agents working continuously.
- Decisions API returns fast constrained model choices.
- Google signs a White House AI self-governance accord.
GPT-6.1 Sol targets cheaper long-running agents
OpenAI has released GPT-6.1 Sol, a model it positions for complex coding, computer use, and multi-step agent work at a much lower price than GPT-6 Astra.
- API pricing is $2 per million input tokens and $10 per million output tokens, with cached input at $0.10 per million tokens.
- OpenAI says cached input is discounted 95% from standard input pricing and costs half as much as GPT-6 Sol cached input.
- On OpenAI’s DeepSWE v1.1 evaluation, Sol scored 75.2% at high reasoning effort versus GPT-6 Sol’s best 68.8%, at approximately 76% lower cost per task.
- It is available in the API, ChatGPT Work, and Codex.
What it means Reprice long-running coding agents around a much cheaper cached-context path, then validate OpenAI’s near-Astra performance claims on your own tasks.
Source: OpenAIDevs
ChatGPT Dots put persistent agents on cloud computers
OpenAI’s new Dots are always-on agents with their own cloud computers, connected apps, and enough context to take on ongoing development work while your laptop is closed.
- A Dot can triage recurring bug reports and feature requests across connected apps, scope failing builds, and prepare code changes for review.
- Developers can give a Dot an ongoing goal, inspect its cloud computer, or connect their own computer for access to local files and code.
- OpenAI says Dots can build and test changes with Codex, then return complete pull requests for review.
- You can set boundaries for autonomous work, approvals, and actions a Dot must never take.
What it means Treat persistent agents as a new product surface for recurring, review-gated work, but keep approvals tight before giving one connected apps or code access.
Source: OpenAIDevs
OpenAI previews a Decisions API for fixed choices
OpenAI is testing a Decisions API powered by GPT-6 Luna for applications that need a fast, constrained choice instead of a generated response.
- You define a question and its possible answers, then use the result to classify content, route a request, or choose an agent’s next action.
- The API accepts text or image context, such as a support request and the teams that could receive it.
- OpenAI describes it as real-time decision-making and says preview access is limited to selected API customers.
- A broader release is planned in the coming days, according to OpenAI.
What it means Move bounded routing and approval decisions out of generative prompts when a fixed choice is enough, while the preview matures into broader API access.
Source: OpenAIDevs
Google signs a White House AI self-governance accord
Google says it signed the White House Accord on Super Intelligence, a proposed industry self-governance effort whose detailed terms and enforcement remain unclear.
- Google said the accord and a Joint Commitment on Frontier Responsibilities create a basis for industry norms and public confidence.
- Google described its own approach as testing, evaluations, red-teaming, and review before releasing models or products.
- Semafor reported that the still-unreleased pact was framed as self-regulation, with discussion of internal controls and internal and external review.
- The available reports do not establish a new enforceable federal AI safety regime or spell out consequences for companies that fall short.
What it means Watch the accord’s released terms, signatories, and enforcement: the signing signals industry coordination, but it is not yet an enforceable federal AI safety regime.
Source: sundarpichai