OpenAI DevDay 2026 Recap: Models, Pricing and Dots to Check

OpenAI announced about 20 things at DevDay on September 29, 2026, and three of them are worth checking today. GPT-6.1 Sol is out (API input is $2 per 1M tokens). Ultrafast, a paid speed tier, is open starting with Astra (OpenAI's price list shows $60 input and $300 output, 6x the standard rate). And there is a new $500 per month (per The Decoder) plan called Pro 500 (as of September 30). I could not confirm availability in Korea or Korean won pricing from official sources.

This article puts the announcements in one table and points out which screens marketers and developers should open today. The source is OpenAI's September 29 post in its developer community, DevDay 2026 announcements and developer resources. I checked the Ultrafast price directly in the OpenAI API price list and developer docs, and cross-checked plan limits against reports from The Decoder and BGR. Each sentence says which source it comes from.

Timing differs by item. GPT-6.1 Sol, Pro 500 and the reopening of new Pro 200 sign-ups start on the announcement day, September 29. The Agents API public beta started on September 11, according to The Decoder. Collaboration slides are only "within weeks," the Private Inference preview is "this fall," and Ultrafast for Sol is just "soon," so I left those dates open. The KRW conversions below assume 1,400 won per dollar. They are estimates, not actual charges.

Read these three lines first

• Model: GPT-6.1 Sol improves coding and computer use, and its standard token price is one fifth of Astra's. The price details are in a separate article, GPT-6.1 Sol price guide.
• Speed: Ultrafast promises up to 8x faster token generation in Codex and up to 6x in the API. For now it works only on Astra, and Sol is "soon."
• Plans: Pro 500 gives 25x Plus usage and Ultrafast. New Pro 200 sign-ups reopened, but the limits differ from the old ones.

How to group the announcements

OpenAI's post splits the announcements into five groups: models and API, Codex, plugins, dots and collaboration, and plans. This article follows the same order. The status column in the table below copies the wording of OpenAI's post, and for prices I list the source I checked. The official pages the post links for each item are Introducing GPT-6.1 Sol, Introducing the Agents API, Introducing dots, the Ultrafast docs, and the ChatGPT Pro tiers help page. Of these, MCPY opened and read only the Ultrafast docs and the API price list. For the rest, I checked only the titles and addresses.

Announcement	What it is	Status and audience	Price and notes
GPT-6.1 Sol	Model with better coding and computer use	Released (Sept 29)	$2 input, $10 output (per 1M tokens). One fifth of Astra
Ultrafast	Paid speed tier	API: open to all users on Astra with low limits, Sol soon. ChatGPT and Codex: Pro 500 and Enterprise (BGR)	Astra $60 input, $300 output (OpenAI price list, 6x standard)
Agents API	Hosted execution, memory, tools, multi-agent	Public beta	Includes computer use (viewing the screen and operating programs)
Decisions API	Classification and routing based on Luna	Limited preview	Price not disclosed. Promises 150ms responses (The Decoder)
Codex Cloud	Cloud tasks that keep running when you close your laptop	Accessible from multiple devices	No price mentioned
Codex Security Cloud	Vulnerability scans, deduplication, fix proposals	Scheduled scans available	The Decoder lists Pro, Business, Enterprise and Edu (not in the post)
dots	Always-on agents with connected apps and their own cloud computer	Rolling out (Axios, BGR)	Text messaging soon, team-level features in planning
ChatGPT Space and Pages	File and document space shared by teams and agents	Available now (collaboration slides within weeks)	Availability by plan confirmed only through press reports
Pro 500	25x Plus usage and Ultrafast	Newly available	$500 per month (The Decoder)

Some announcements are not in the table. These are calling @ChatGPT in Slack and Teams, a macOS beta meeting notes plugin, Sign in with ChatGPT for logging in to other apps (16 launch partners), an enterprise marketplace, and Managed Agents for running OpenAI agents on Amazon Bedrock. I left them out because their practical impact depends on your account and contract.

What changed in models and the API

The API model ID for GPT-6.1 Sol is gpt-6.1-sol, at $2 per 1M input tokens and $10 output. A token is the unit a model uses to read and write text. Input and output prices match the existing GPT-6 Sol, and only cached input got cheaper. Detailed calculations and KRW examples are in GPT-6.1 Sol price and Astra comparison, so I won't repeat them here.

Ultrafast is a pricing option, not a model. It runs the same model faster in exchange for a higher unit price. OpenAI says up to 8x in Codex and up to 6x in the API. The Ultrafast entry in the API price list, for gpt-6-astra, is $60 per 1M input tokens, $6 cached input and $300 output. That is 6x the standard rates ($10 input, $50 output). Long requests with more than 272,000 input tokens go up to $120 input and $450 output. At 1,400 won per dollar, 1M input tokens cost 84,000 won and 1M output tokens cost 420,000 won. Check the exchange rate and VAT again on your billing screen.

To use it in the API, set model to gpt-6-astra and service_tier to ultrafast in the request. The developer docs say Ultrafast for Astra is open to all API users with low request limits. For agents that make many tool calls, the docs recommend a WebSocket connection, because opening a new connection every time reduces the latency gain.

The Agents API is a public beta. OpenAI handles the server where agents run, conversation memory, tool calls and the division of work among multiple agents. The API also includes a computer use tool. It operates programs the way a person would, by looking at the screen and clicking. If your team has been building workflow automation, it is time to ask whether moving your self-built execution server to this API is worth it. The post has no price, so I could not confirm one.

On the security side, Private Intelligence stands out. It adds Private Safety Processing to the ZDR (zero data retention) option, which does not store inputs. A stronger Private Inference preview is planned for this fall. Teams that handle customer data should check the terms of availability and Korean personal data outsourcing requirements in their contract.

What is new in Codex, dots and Spaces

Codex changed in four ways. Codex Cloud keeps tasks running even after you close your laptop. The CLI gained two-way voice and an /agents screen. Code Review puts change summaries and diff questions into the review screen and runs automatic reviews in the cloud. Codex Security Cloud scans repositories for vulnerabilities, removes duplicates and even prepares fixes. The Decoder lists Codex Security Cloud for Pro, Business, Enterprise and Edu, but OpenAI's post does not name plans, so I do not treat that as confirmed.

dots is the headline of this event. An always-on agent stays running even when I don't ask anything, and keeps working with connected apps and its own cloud computer. Controlling it by text message is "soon," and multiple dots working as a team is in the planning stage. Enterprise has a preview of specialized dots with defined roles, and integration with Microsoft Agent 365 is planned.

Reports differ on who can use it. BGR lists Pro 200 and above, Business Premium and Enterprise, while another report lists Pro and Business Premium. Neither includes Plus. No regional restrictions are mentioned anywhere. MCPY marks this cell as "varies by plan, check the official help page."

ChatGPT Space is where teams and dots share the same files and project context, and Pages are documents that teammates and agents edit together there. They can even include charts and interactive tools. Collaboration slides arrive within weeks.

What changed in the Pro plans

Pro 500 costs $500 per month and gives 25x Plus usage and Ultrafast. According to BGR, Pro 200 stopped taking new sign-ups on September 10 and reopened on September 29, but new subscribers get different limits than existing ones. BGR and The Decoder both reported the same numbers: new Pro 200 Work and Codex usage dropped from 20x Plus to 10x. These numbers are based on those two outlets.

MCPY left the weekly GPT-6 Pro message limit for Pro 200 out of the table. The figure appeared in only one outlet, and I could not match it against another source. People already on Pro 200 will most want to know whether their limits stay the same, but the post only says that new subscribers get different limits.

What to check in Korea

• Won pricing: Pro 500 is $500 per month in dollars. At 1,400 won per dollar that is 700,000 won, but the amount on the Korean checkout screen and VAT are unconfirmed. Korean won and easy-pay coverage for ChatGPT subscriptions is covered in ChatGPT Korea won payment guide.
• Availability by region: none of dots, Spaces or Ultrafast mention regional limits. No mention does not mean it works in Korea. Check whether it appears in your own account.
• Data outsourcing: before connecting customer data to dots or the Agents API, check the retention period and Korean personal data processing outsourcing terms.
• Ad account connections: when a marketing team connects ad and analytics accounts to dots, start with read-only permissions.

Common mistakes

The most common mistake is reading Ultrafast as a new model. The model is the same, a speed option is added, and for now it works only on Astra. Second, people multiply "Sol is one fifth of Astra" by "Ultrafast is 6x" to guess the Sol Ultrafast price. Sol Ultrafast is not out yet and has no price. Third, people apply the new Pro 200 limits to existing subscribers. Fourth, people tie beta and preview features (Agents API, Decisions API, specialized dots) directly to production work. Until limits and prices are set, it is safer to run them in parallel with your existing flow and only compare results.

The order to check things today

1. On the chatgpt.com plans screen, see whether your account shows Plus, Pro 200 or Pro 500, and whether the limit wording has changed.
2. Check whether GPT-6.1 Sol is in the model picker for Work and Codex.
3. If you use the API, change the model ID to gpt-6.1-sol and run the same three requests side by side with Astra.
4. If you test Ultrafast in the API, set model to gpt-6-astra and service_tier to ultrafast, make just one call, and check that the unit price on the usage screen matches the price list ($60 input, $300 output). Request limits are low, so do not make bulk calls.

This was right after the announcement, so limits and availability may change within days. Open the OpenAI plans screen once more before you pay.

Sources

• OpenAI developer community, DevDay 2026 announcements and developer resources (September 29, 2026)
• OpenAI, DevDay 2026 Recap (not read directly)
• OpenAI developer docs, Ultrafast mode
• OpenAI developer docs, API price list (Ultrafast price entry)
• OpenAI, Introducing GPT-6.1 Sol
• OpenAI, Introducing the Agents API
• OpenAI, Introducing dots
• OpenAI Help, About ChatGPT Pro tiers
• The Decoder, OpenAI DevDay recap
• BGR, Everything OpenAI Announced At DevDay 2026
• CNBC, OpenAI DevDay 2026 live updates