This Week's Releases: OpenAI's Dots and GPT-6.1 Sol, Claude Sonnet 5.5 and Meta's Enterprise Platform
Title: This Week's Releases: OpenAI's Dots and GPT-6.1 Sol, Claude Sonnet 5.5 and Meta's Enterprise Platform
Date: 30 September 2026
Type: Blog
Author: SAASiQ (contact@saasiq.ai)
Word count: 1587 words
Reading time: 6 min
Published: 30-09-2026
OpenAI launched always-on agents called dots at its DevDay conference in San Francisco on 29 September. OpenAI also released GPT-6.1 Sol, a cheaper model that it says comes close to its flagship GPT-6 Astra. Anthropic released Claude Sonnet 5.5 on 28 September at the same price as Sonnet 5. Meta announced an enterprise platform for its Muse agent and models on the same day. Nvidia and more than 100 partners launched an open platform for keeping AI agents within set limits. The US government opened America.gov, a chatbot for federal services, on 29 September.
Dots: agents that keep running
Dots were the main launch at DevDay. A dot is an agent that runs in OpenAI's cloud on GPT-6 Astra, OpenAI's most capable released model. Each dot gets its own cloud computer and web browser. It keeps working on the goals it is given after the user closes the laptop. OpenAI calls them "remarkably capable, always-on agents".
A dot can connect to more than 4,000 apps through plugins. Users can reach it in ChatGPT, on a voice call, or in Slack and Microsoft Teams, and OpenAI says texting is coming soon. The first dot is included in the Pro and Business Premium plans in eligible markets. Conversations with it do not count towards ChatGPT usage limits. Enterprise, Edu and Healthcare workspaces get dots as a beta, switched off by default, so an administrator decides whether they can reach company data.
OpenAI also launched ChatGPT Space, a shared workspace where team members and their dots work on the same files. Space replaces the existing ChatGPT Library for Pro, Business and Enterprise users. It holds Pages, a new kind of document with charts, images, checklists and dashboards that people and dots edit together. Plugin extensions let developers build an editor or a dashboard directly into ChatGPT and Codex.
GPT-6 Astra is the model the UK AI Security Institute tested before its release. In those simulated tests, published on 28 September, it carried out a supply-chain attack in 29.2 per cent of runs with its cyber safeguards switched off. SAASiQ covered the results on 29 September.
GPT-6.1 Sol, Ultrafast and Pro 500
GPT-6.1 Sol replaces GPT-6 Sol, which OpenAI released on 22 September. It keeps the same price of $2 per million input tokens and $10 per million output tokens. Cached input, meaning material sent over and over such as a standing set of instructions, now costs $0.10 per million tokens. That is half of GPT-6 Sol's cached price.
OpenAI says the model nearly matches GPT-6 Astra on agentic coding, computer use and professional work, at a fifth of Astra's price. Astra costs $10 and $50. On Terminal-Bench Science 0.1, a test of scientific work such as data analysis and simulation, OpenAI reports that GPT-6.1 Sol more than doubles GPT-6 Sol's score. It averaged $5.47 per task on that test. These are OpenAI's own figures.
Ultrafast is a new speed tier. OpenAI says it generates text up to eight times faster in Codex, at 300 tokens a second, and up to six times faster in the API. In the API it costs six times the standard price. GPT-6 Astra Ultrafast is available now in the API and on the Pro 500 and Enterprise plans. GPT-6.1 Sol Ultrafast is listed as coming soon.
Pro 500 is a new ChatGPT plan at $500 a month. It carries 25 times the usage allowance of ChatGPT Plus and is the only Pro plan with Ultrafast. Codex, OpenAI's coding assistant, can now run tasks in the cloud, so a user can start one from a phone. Its command-line tool also accepts spoken instructions.
Claude Sonnet 5.5
Anthropic released Claude Sonnet 5.5 on 28 September, six days after Claude Opus 5.5. It keeps Sonnet 5's price of $2 per million input tokens and $10 per million output tokens, the same list price as GPT-6.1 Sol. Anthropic says it produces output more than 30 per cent faster than Sonnet 5. It also says the model costs up to 30 per cent less per task in its own testing.
Anthropic's figures put it close to Opus 5.5, which costs twice as much. On Terminal-Bench 4.0, a test of agents working at the command line, Sonnet 5.5 scores 70.6 per cent. Opus 5.5 scores 66.4 and Sonnet 5 scored 10.3. On CursorBench 4.0, a coding test, Sonnet 5.5 scores 55.5 per cent against 57.8 for Opus 5.5.
Artificial Analysis, which tests models independently, ranks Sonnet 5.5 second on its Intelligence Index. It also found that at maximum effort the model wrote about 193,000 tokens per test task. That is the most Artificial Analysis has measured, and roughly 60 per cent more than Opus 5.5. It came to $7.60 per task, about 50 per cent above Sonnet 5, which runs against Anthropic's claim of savings.
Sonnet 5.5 is on the free Claude plan as well as the paid ones. It has a context window of one million tokens, so it can take in very long documents in one go. It is in Anthropic's API and on Amazon Bedrock, Google Cloud's Vertex AI and Microsoft Azure.
Meta's enterprise platform and Muse for Small Business
Meta announced the Meta Enterprise Platform on 28 September. It packages four products for businesses and developers: the Muse agent, Meta Business Agent, the Muse API and Muse Code. Meta has hired Chirantan "CJ" Desai, chief executive of the database company MongoDB, to run it. He becomes chief enterprise platform officer and reports to Mark Zuckerberg.
The announcement had no enterprise prices, launch dates, named customers or regions. Some parts already exist. Muse Code has been in beta since August. The Muse Spark API has been priced since July, at $1.25 per million input tokens and $4.25 per million output tokens for the standard tier.
On 29 September Meta launched Muse for Small Business in the US and Canada. It connects to services including Shopify, QuickBooks, Stripe, Slack, Asana, Notion and Zoom, as well as Facebook and Instagram business accounts. It is free with usage limits. Two paid plans add more: 500 million tokens a week for $20 a month, or 3 billion tokens a week for $100 a month.
Nvidia's Open Agent Safety Platform
Nvidia launched the Open Agent Safety Platform on 28 September with more than 100 backers. They include Anthropic, Microsoft, Salesforce, Cisco, CrowdStrike, Hugging Face and JPMorganChase. The platform is software plus a reference hardware design. Its purpose is to keep AI agents within the limits set for them, from testing through to live use.
It has two parts. OpenShell is an open-source runtime, the environment an agent runs inside, released under the Apache 2.0 licence. It is available now on GitHub and can run on Arm and Intel hardware as well as Nvidia's. Sentry is a watchdog that runs on Nvidia's BlueField-4 chips, separate from the agent. Nvidia says it can quarantine an agent within milliseconds if the agent tries to go beyond its permitted limits.
Sentry is a reference design with no availability date. BlueField-4 is expected in early availability with Nvidia's Vera Rubin systems in 2026.
America.gov
The US government launched America.gov on Tuesday 29 September. It is a chatbot that answers questions about federal services, drawing on about 29,000 federal websites. The US Chief Design Officer, Joe Gebbia, said it runs on Google's Gemini and xAI's Grok. Users can type or speak their questions, and no account is needed.
For now the site finds information. The administration says it plans to let people start or complete tasks through it later, such as renewing a passport, enrolling in Medicare, checking student-loan details and reserving federal campsites.
Smaller releases
Google began testing Call for Me on 24 September with a small group of Pixel 11 owners in the US. Gemini phones a business on the user's behalf to book an appointment or change a reservation. It introduces itself, waits on hold and works through automated phone menus. The user sees a live transcript and can take over at any point. The test is in English only and needs a paid Gemini subscription. Google has given no date for a wider rollout.
Alibaba's Qwen team released Qwen3.8 Max Prime on 23 September. It is a faster version of Qwen3.8 Max, with 1.5 to 2 times the output speed, sold separately at $4 per million input tokens and $12 per million output tokens. It has a context window of one million tokens and accepts text, images and video.
Before switching
Most of the performance figures above come from the companies that make the models. The one independent cost test this week, Artificial Analysis on Sonnet 5.5, found a higher cost per task than Anthropic's own. GPT-6.1 Sol and Sonnet 5.5 now share a list price of $2 and $10. So for many teams the difference will come down to how many tokens each model uses on their own work. A sample of real tasks run through both gives the cost per completed task, which is the figure to compare.
Dots, Meta's Muse agent and Microsoft's Autopilot, announced on 25 September, are all agents that keep running in the supplier's cloud. Before switching one on, an administrator can check which apps it can reach, where its activity is logged and who can stop it.
GPT-6.1 Sol Ultrafast is listed as coming soon, and Meta has not given dates for its enterprise products. OpenAI's Sam Altman and Anthropic's Dario Amodei have been asked to appear before the Australian Senate inquiry in Canberra on Thursday 1 October.
SAASiQ - Intelligent Solutions for SaaS ©
