OpenAI DevDay 2026: GPT-6.1 Sol Pricing, Agents API and Microsoft Teams Integration

What OpenAI announced​

OpenAI announced more than 20 updates across its models, ChatGPT, Codex, APIs and enterprise products during DevDay 2026. The headline items were GPT-6.1 Sol, the Dots agents, ChatGPT Space and a $500 Pro 500 tier.

The weekly roundup from Patrick McGuinness lists these items. Where OpenAI's own documentation or other coverage was available, this article uses that instead. Benchmark claims are OpenAI-reported. One roundup notes they are not independent head-to-head tests and can differ from production ChatGPT.

GPT-6.1 Sol: pricing and availability​

OpenAI's API changelog for September 29 lists the model as gpt-6.1-sol. It is aimed at complex coding and professional work at a lower cost than GPT-6 Astra. The changelog gives standard pricing for prompts up to 272K input tokens:

Longer prompts are priced differently. OpenAI's documentation says prompts over 272,000 input tokens double the input and cache rates. Output costs 1.5 times the standard rate for the full request. Budget for that if you feed whole repositories into one request.

Availability differs by product:

• API: The changelog says to use the Responses API for tool calling. It also notes beta support for multi-agent delegation, where the model hands work to subagents within a single request.
• ChatGPT: OpenAI's help page says Sol is rolling out in ChatGPT Work and Codex. If you don't see it, it may not be available for your account yet. One report says it is not yet in regular chat.
• Enterprise and Edu: OpenAI's help page says workspace owners must enable model access. Admins should expect it to be off until they turn it on.
• Ultrafast: This is a processing tier, not a separate model. It was announced starting with Astra, and OpenAI said Sol support comes later. The "up to 300 tokens per second" figure in the roundup does match coverage that cites 300 tokens per second in Codex for supported workloads. Treat it as a vendor figure.

Agents API and computer use​

The Agents API gained computer use on September 29. According to the changelog, agents can complete tasks in an OpenAI-hosted browser. Website access approvals and sign-in are handled by your application.

For administrators, this is a hosted-browser capability. It is not a local desktop agent, and the calling application decides which sites and credentials the agent can reach. The changelog also records that the Decisions API is Luna-powered and aimed at tasks such as classification and routing.

Microsoft Teams integration​

The part most relevant to Microsoft 365 shops is that @ChatGPT now works in Slack and Microsoft Teams. One roundup lists it as live for Business and Enterprise. Another says coworkers can add context and follow-up requests in a channel, thread or direct message. An individual ChatGPT license isn't required in enabled channels, and organization settings and connected accounts determine access.

This is an OpenAI integration with Teams, not a Microsoft feature. Before enabling it, check these points:

1. Which app permissions your tenant grants the integration.
2. Which connected tools and accounts the bot can use in a channel.
3. Whether channel members who have no ChatGPT license should see output drawn from connected data.
4. How you will review agent actions that touch shared files.

Dots and ChatGPT Space​

Each dot runs on GPT-6 Astra and has its own cloud computer, and it can connect to more than 4,000 apps. Rollout is staged. One tracker lists Pro and Business Premium in eligible markets, with an Enterprise, Edu and Healthcare beta when an admin enables it.

The roundup's "24/7" framing is broadly consistent with OpenAI's documentation. That documentation says a dot keeps working in the cloud while the user's laptop is off. Claims about reliability are less settled. The roundup mentions bugs in this first release, but I found no primary-source testing either way.

ChatGPT Space is described as a collaborative space where teammates, agents, documents, plugins and recurring tasks work together. A tracker lists it as live on desktop and web for Pro, Business and Enterprise.

Pricing tiers​

Pro 500 costs $500 a month with 25 times the Plus allowance and Astra Ultrafast access. Coverage also describes a usage cut for Pro 200. Don't confuse ChatGPT subscription tiers with the API service_tier setting. Ultrafast exists in both places, but under different terms.

Codex Security Cloud​

OpenAI's developer post on Daybreak, dated August 21, describes Codex Security Cloud as available in research preview. It says the service had analyzed more than 30 million commits across more than 30,000 codebases. That figure is OpenAI-reported.

The workflow is review-first:

1. Connect a GitHub repository.
2. Choose a one-off repository scan or ongoing commit monitoring.
3. Review the evidence and the validation output.
4. Inspect any proposed patch before opening a pull request.

OpenAI says a larger repository scan may take several hours. It also warns that local CLI scans use your operating-system permissions without pausing for approval. Remove unrelated credentials from the environment, and keep reports private because they can contain source excerpts. Findings posted to GitHub inherit the pull request's visibility.

The other frontier releases this week​

Anthropic Claude Sonnet 5.5. Anthropic prices it at $2 per million input tokens and $10 per million output tokens. Anthropic's page says it matches GPT-6 Sol's best FrontierCode score at High effort for about a fifth of the cost per task. That is a company-reported comparison. The page also says it is available on Azure, AWS and Google Cloud. It adds that higher-risk cybersecurity tasks will visibly fall back to Sonnet 5.

Google Gemini 4 Argon. Google's October 1 announcement says it is rolling out first to trusted cyber defenders through the Fairwind Program. Broader release to developers, enterprises and consumers is planned, with paid API customers and Google AI Ultra subscribers first. No date was given. The announced introductory price is $2 per million input tokens and $10 per million output tokens. Output limits rise to 1M tokens from 64K.

On list price, then, all three vendors land at $2 and $10. The real differences are token efficiency, access and safeguards.

Security context​

OpenAI's system-card addendum rates GPT-6.1 Sol "Critical" for cybersecurity capability, per the research brief. It says Sol uses the same safeguards stack as GPT-6 Astra. The roundup separately reports that OpenAI notified more than 100 outside organizations after a review of the Hugging Face security incident. Notification did not necessarily mean a compromise, and OpenAI withheld many details. California's attorney general has subpoenaed OpenAI over the matter, according to the roundup.

Nvidia's open-source Open Agent Safety Platform, covered in the roundup, targets the same problem. It aims to run agents in sandboxes with kernel-level isolation and policy enforcement over file, network and tool access. I haven't verified it independently.

Other items in the roundup​

These are secondary for Windows and IT readers:

• AMD agreed to acquire World Labs for about $8.2 billion in an all-stock deal, expected to close by the end of 2026 subject to approvals.
• A voluntary White House accord on frontier model safety calls for internal and external reviews. It is voluntary, not binding regulation.
• Other items include Google's Gemini Live Guided Vision, xAI Team Bots, Suno Speech, Meta's Muse hardware code and Amazon's Strands Decider 2B. None is a Microsoft development.

What to do now​

• Developers: Benchmark GPT-6.1 Sol against Sonnet 5.5 on your own tasks. Model the long-prompt pricing before migrating.
• Teams admins: Hold the @ChatGPT integration until app permissions and connected tools are reviewed.
• Security teams: Pilot Codex Security on a non-critical repository. Keep human review on every patch.
• Budget owners: Treat Ultrafast and Pro 500 as premium speed purchases, not defaults.

The price war is real, but at this point it is a three-way tie on list price. The practical question is which model finishes your task in the fewest tokens, with the permissions you are willing to grant it.