01Models & economicsSEPT 29
GPT-6.1 Sol arrives with cheaper cached context
INPUT$2per 1M tokens
CACHED INPUT$0.10per 1M tokens
OUTPUT$10per 1M tokens
GPT-6.1 Sol · Standard API pricing, per OpenAI
Announced September 29. OpenAI released GPT-6.1 Sol in the API, ChatGPT Work, and Codex. Standard API pricing is $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens. The API model identifier is gpt-6.1-sol. OpenAI says it approaches GPT-6 Astra on several coding, computer-use, and professional-work evaluations at one-fifth of Astra's standard input and output token prices. Those are vendor evaluations, not independently reproduced results. Availability includes eligible paid ChatGPT plans in Work and Codex; the announcement specifically says it is not yet available in Chat. OpenAI's model announcement ↗ [1]
WHY IT MATTERS Our readAgents that repeatedly consult the same documents or codebase can benefit from cheaper cached context. The useful comparison is quality and total cost on your own task set, including failed attempts, rather than a single benchmark score.
02Personal agentsSEPT 29
Dots turns OpenAI's agent pitch into an ongoing relationship
Announced September 29. OpenAI introduced Dots, agents powered by GPT-6 Astra with their own cloud computers and access to connected apps. Rollout begins on Pro and Business Premium in eligible markets; enterprise users can try a beta when their administrator enables it. An important boundary: OpenAI says the proactive research that runs when users are not actively working with a dot uses read-only tools. That mode cannot send messages, change connected-app content, or control a browser or computer. Directed tasks have separate action and approval controls. OpenAI's Dots announcement ↗ [2]
WHY IT MATTERS Our readThe product is moving from answering a prompt to retaining responsibility for work. That makes visible activity, clear access boundaries, and useful handoffs central to the experience. Always available should not be confused with unlimited permission to act.
03Shared knowledgeSEPT 29
ChatGPT adds shared workspaces, documents, and richer plugins
Announced September 29. At DevDay, OpenAI introduced ChatGPT Space for teams and agents to build on shared knowledge, and Pages for collaborative documents. Space is available on Pro, Business, and Enterprise on desktop and web; mobile supports finding, reading, and sharing pages, with creation and editing still to come. Separately, plugin extensions let developers add sidebar destinations, interactive panels, and file viewers inside ChatGPT. OpenAI's DevDay recap ↗ [3]
WHY IT MATTERS Our readThis expands the competition beyond chat interfaces into the places where teams keep their work. Developers gain another distribution surface, but should consider how customers export documents, retain permissions, and keep context usable across assistants. Integration depth and portability will increasingly need to coexist.
04Small businessSEPT 29
Meta expands Muse into small-business operations
Announced September 29. Meta's Muse for Small Business adds skills and connectors to its existing agent, which is available in the US and Canada. The expansion connects Instagram professional-account analytics, Facebook Pages, and Meta ad accounts alongside services such as Shopify, QuickBooks, Canva, and Slack. Meta describes workflows spanning campaign planning, sales analysis, and financial reviews. It says publishing, sending, and spending require the user's approval. This is a concrete product expansion following the business-platform organization announcement covered yesterday. Meta's announcement ↗ [4]
WHY IT MATTERS Our readSmall businesses rarely need another empty chat window. They need help with the tools and records they already have. The test is whether connecting those systems produces dependable work with less supervision, especially when money or customer communications are involved.
05Policy & oversightSEPT 29
The White House secures a voluntary AI safety accord
Announced September 29. The White House and leaders from Google, Meta, Anthropic, SpaceXAI, NVIDIA, and OpenAI signed an accord addressing risks from advanced AI. The published text calls for internal controls and monitoring, internal review teams, independent external evaluations, and oversight by an independent board committee. It also calls for continued work on standards. These are commitments to an oversight process, not evidence that every safeguard is already operating. Accord text reproduced by the Washington Examiner ↗ [5]
AFP's reporting describes the agreement as voluntary and not legally enforceable. AFP's report ↗ [6]
WHY IT MATTERS Our readExternal evaluation could make safety claims more inspectable. The next questions are who performs the evaluations, what gets disclosed, and what happens when controls fail. A signed pledge alone does not answer them.