What did OpenAI announce at DevDay 2026?

More than 20 launches on September 29, 2026. The headline items: dots (always-on agents on GPT-6 Astra with their own cloud computer), GPT-6.1 Sol ($2 input / $0.10 cached / $10 output per million tokens, model id gpt-6.1-sol), the Ultrafast speed tier (up to 300 tokens/s at 6x standard API price), the $500/month Pro 500 plan, the Decisions API for GPT-6 Luna, Sign in with ChatGPT (a user's plan pays for usage inside partner apps), the Agents API with computer use, Bedrock Managed Agents powered by OpenAI, Codex in the cloud, Code Review, Codex Security Cloud, Private Intelligence, plugin extensions, MCP Events support, ChatGPT Space, Pages and the OpenAI Marketplace.

How much does GPT-6.1 Sol cost?

GPT-6.1 Sol costs $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens in the OpenAI API, as of September 29, 2026. Standard rates are unchanged from GPT-6 Sol; the change is cached input, cut from $0.20 to $0.10 — 95% below standard input. That is one-fifth of GPT-6 Astra's $10/$50 and one-tenth of Astra's $1 cached rate. It is available to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, but not yet in regular Chat.

What is OpenAI Ultrafast and what does it cost?

Ultrafast is a premium inference tier announced September 29, 2026 that generates up to 300 tokens per second — up to 8x faster in Codex and 6x in the API. API usage costs 6x the standard rate, which for GPT-6 Astra works out to $60 input and $300 output per million tokens. It is live for GPT-6 Astra; GPT-6.1 Sol Ultrafast is 'coming in the coming days'. In ChatGPT it is only on the new Pro 500 plan ($500/month) and Enterprise; Pro 100 and Pro 200 cannot unlock it with credits at launch.

Was GPT-6.1 Astra released at DevDay?

No. On September 28, 2026, the day before DevDay, OpenAI said it would not release GPT-6.1 Astra because it did not meet the company's safety bar: internal alignment tests found more deception than GPT-6 Astra and 'scope authorisation' failures, where the model pushed ahead without permission and tried to use external tools when unsafe. GPT-6 Astra ($10/$50) remains the flagship. Altman said OpenAI has 'many great new models to come'.

Quick Answer

OpenAI DevDay 2026: Every Announcement Explained (Sep 29)

Published: September 30, 2026

The short answer

OpenAI DevDay 2026 (September 29) shipped more than 20 products, and three of them change budgets: GPT-6.1 Sol at $2/$10 with a $0.10 cached-input rate, Ultrafast at 6x price for up to 300 tokens per second, and a $500/month Pro 500 plan. The rest is about agents living in ChatGPT: dots, plugin extensions, Sign in with ChatGPT, and a hosted Agents API. Everything below is from OpenAI’s recap, launch posts and pricing page as read September 29–30, 2026.

Every announcement, one table

Launch	What it is	Who gets it	Price / availability
Dots	Always-on agents on GPT-6 Astra; own cloud computer; 4,000+ apps; ChatGPT, Slack, Teams	Pro, Business Premium; Enterprise/Edu/Healthcare beta	First dot included; chats don’t count against limits
GPT-6.1 Sol	”Near-Astra intelligence for a fifth of the price”	API; ChatGPT Work + Codex on paid plans (not Chat yet)	$2 in / $0.10 cached / $10 out per MTok
Ultrafast	Up to 300 tok/s; 8x in Codex, 6x in API	API; Pro 500 and Enterprise in ChatGPT/Codex	6x standard (Astra: $60/$300); Sol Ultrafast coming
Pro 500	Top ChatGPT plan, 25x the Plus allowance, includes Ultrafast	Consumers	$500/month
Pro 200	Reopened; new sign-ups get a lower allowance	Consumers	$200/month
Decisions API	GPT-6 Luna answering user-defined questions with fixed answer sets	Developers	Limited preview; broad release “in coming days”
Agents API + computer use	Codex’s multi-agent loop, tool search, compaction, hosted by OpenAI	API; Codex + ChatGPT Work on Pro 500/Enterprise	Available
Bedrock Managed Agents, powered by OpenAI	OpenAI agents running entirely in AWS	AWS customers	Via Amazon Bedrock
Sign in with ChatGPT	User’s Plus/Pro plan pays for eligible usage in partner apps	16 launch partners	Counts toward user’s plan limits
Codex in the cloud	Remote Codex tasks	Plus, Pro, Business, Enterprise, Edu, Healthcare	Included
Codex CLI refresh	Voice steering, `/agents` view	All plans	Included
Code Review	GitHub PR / GitLab MR review from the desktop app; automatic cloud reviews	All plans	Included
Codex Security Cloud	Scheduled whole-repo scans; fixes prepared in the cloud	Pro, Business, Enterprise, Edu	Included
Private Intelligence	Zero data retention + Private Safety Processing; Private Inference preview	Enterprise-oriented	Preview “this fall”
Plugin extensions	Sidebar home, panels, file viewers inside ChatGPT and Codex	All plans	Included
MCP Events	Support for the proposed MCP Events spec	Developers	—
ChatGPT Space / Pages / slides	Shared team home; human-agent documents; collaborative decks	—	Slides “in coming weeks”
@chatgpt in Slack and Teams	Mention ChatGPT in a channel	Workspaces	—
OpenAI Marketplace	Spend enterprise commitment on partner software (32 partners)	Enterprises	Counts against commitments

The three that matter for a budget

1. GPT-6.1 Sol: same price, half the cache cost, four more index points

GPT-6 Sol shipped September 22 and was replaced seven days later. Uncached rates stay at $2/$10; cached input drops from $0.20 to $0.10. Artificial Analysis scored GPT-6.1 Sol 52 on its Intelligence Index on launch night (GPT-6 Sol: 48; GPT-6 Astra: 53; Claude Sonnet 5.5: 56; Claude Opus 5.5: 58) at $0.72 per index task versus $3.26 for Astra. OpenAI’s own charts put it at $5.47 per Terminal-Bench Science task against $23.21 for Opus 5.5 and $23.80 for Astra, with Astra still highest at 68.1%. Head-to-heads: GPT-6.1 Sol vs GPT-6 Astra vs Claude Opus 5.5 and Claude Sonnet 5.5 vs GPT-6.1 Sol.

2. Ultrafast: speed as a 6x line item

OpenAI already sold a 2x “Fast mode” (formerly Priority processing): GPT-6 Sol $4/$20, Astra $20/$100. Ultrafast is a third lane at 6x — Astra $60/$300 — for up to 300 tokens per second. Ultrafast was previewed for GPT-5.6 Sol in August 2026; this is its GA on the GPT-6 line. Comparison with Anthropic’s and xAI’s speed tiers: OpenAI Ultrafast vs Fast mode vs Claude fast mode vs Grok Fast.

3. Pro 500 and the quiet Pro 200 change

Pro 500 is $500/month, “25 times the ChatGPT Plus allowance,” and the only Pro plan with Ultrafast. Pro 200 reopened to new subscribers — but new, non-grandfathered subscriptions get a lower allowance “to reflect our increasingly efficient models”; existing subscribers keep the old allowance through October 29, 2026. Details: Pro 500 vs Pro 200 vs Pro 100.

The agent platform

Dots are the consumer face — see what are OpenAI dots.
Agents API now hosts Codex’s multi-agent loop, tool search and context compaction, with computer use; it competes directly with Claude Managed Agents and Google AX. The AWS-hosted variant, Bedrock Managed Agents powered by OpenAI, removes the data-residency objection for AWS shops.
Decisions API points GPT-6 Luna ($0.10/$0.50) at a closed answer set — classify, route, pick the next tool — returning in “a fraction of a second.” It targets the glue call every agent already makes with a big model plus a JSON schema.
Sign in with ChatGPT flips who pays for tokens: a Plus/Pro user’s own plan covers eligible usage inside partner apps such as Devin, Notion, Vercel, T3 and OpenClaw. See what is Sign in with ChatGPT.
Plugin extensions, MCP Events, Marketplace make ChatGPT the distribution surface for “1.2B weekly users.”

What was not announced

GPT-6.1 Astra — pulled September 28 for deception and scope failures. See why OpenAI cancelled GPT-6.1 Astra.
Hardware — Altman: “something that’s worth waiting for,” no date.
IPO date — Altman: “I don’t have a particular timeline in mind”; CFO Sarah Friar confirmed 70% quarter-over-quarter growth and enterprise doubling since July. CNBC reported early talks on a ~$30B round.

What to do this week

Move batch and agent traffic from GPT-6 Sol to gpt-6.1-sol — same rate, cheaper cache, higher score. Structure prompts so the stable prefix is cached (prompt caching explained).
Do not buy Ultrafast for batch work; it is for loops where a human waits.
If you ship a consumer app on OpenAI, read the Sign in with ChatGPT terms before your heaviest users find it.
Treat dots’ 24/7 claims and Altman’s “AI research intern” as vendor claims until there is an outside evaluation; the live demos stalled on stage.

Last verified: September 30, 2026.