The Most Important New AI Tools from OpenAI Dev Day

September 30, 2026 · Episode Links & Takeaways

Author

AI Daily Brief
September 30, 2026

MAIN STORY

The Biggest AI Announcements from OpenAI Dev Day

More than 20 launches landed at Dev Day, many arriving far earlier than planned because Astra-class models have dramatically sped up OpenAI's own development. None of it rewrites the industry's trajectory. Instead, it confirms the trends that matter most: cheaper good-enough models, persistent always-on agents, and AI work moving from single player to multiplayer.

OPENAI ACCELERATES ON DEV DAY

Dots
OpenAI's answer to Muse and GrokBot, piloted by Astra
Persistent, always-on agents with their own cloud computer, reachable by text, voice call, Slack and Teams, but capped at one per user and limited to Pro, Business and Enterprise for now, a real handicap against a free Muse. The demo stumble matters less than the teething bugs early users hit; the product leans more Codex orchestrator than personal assistant, and it's a space to keep watching.

Decisions API
Judgment models are becoming a fundamental primitive
A version of Luna that mimics Jev's fast classification for routing, labeling and choosing an agent's next action, claimed at 10x faster than the Responses API, and with visual inputs Jev can't take natively. It's only in limited preview, but the real signal is that every frontier lab will likely ship a judgment model to sit alongside its generative one.

Space
The sleeper hit power users loved right away
A shared, AI-native workspace for docs, sheets, slides and scheduled automations, where Dots work natively alongside human teams. It reads as a shot at Microsoft 365, Google Drive and Notion, but those tools have been a bad fit for agentic work, and a truly AI-native suite was only a matter of time.

GPT-6.1 Sol
"Near-Astra intelligence for a fifth of the price"
Just a week after 6 Sol, 6.1 Sol on medium beats 6 Sol on max, edges Astra on DeepSWE at 75.2% (high), and scores 71.4% on OSWorld at a third of 6 Sol's cost. Artificial Analysis put it one point shy of Astra at a quarter of the cost, though first hands-on reactions are sharply split.

No GPT-6.1 Astra
Sol-class models may have to be good enough
OpenAI scrapped the release over safety concerns after failing to dial down the model's tenacity without making it lazy. Head of safety systems Saachi Jain said it "didn't quite meet the bar in terms of staying within scope and authorization," a sign the next frontier steps need technical breakthroughs first.

Plugin Extensions
Could ChatGPT finally become an app store?
Developers can now ship full native apps inside ChatGPT, with relevant plugins surfaced in conversation to 1.2B+ weekly users, after more than a year of OpenAI testing plugins, integrations and MCP. Builders are already rushing in, while skeptics question whether handing OpenAI the user traces is worth it.

Sign in with ChatGPT
Carry your intelligence subscription into any app
Third-party apps can let users sign in with ChatGPT and route API costs through the existing subscription, so apps price only the app layer. Losing a markup on tokens is likely worth never having to convince users to pay twice for intelligence.

Ultra Fast Mode and the $500 Tier
"Compute constraints are real, present, and permanent"
Ultra Fast Mode delivers 8x faster tokens in Codex and 6x in the app, for now only on Astra and only on a new $500 tier with 25x Plus usage. The $200 Pro tier also reopened at roughly half its previous value, a painful trade that should force a longer-term push toward efficiency.

Codex Cloud Environments
Keep working after the laptop closes
Codex gets a dedicated cloud environment, part of the industry-wide shift since GrokBot toward cloud instances as a standard piece of every agentic product, and AI as something persistent and always on.

Codex CLI
A win for the monothread maxis
A refreshed CLI with deeper scrollback history, a pinned composer, and a new /agents command for checking on and jumping between parallel tasks.

Private Intelligence
Zero data retention, even at inference time
With growing fears that labs skim customer data, a theme running through nearly all of Microsoft's messaging lately, a clear end-to-end privacy guarantee is a big deal for enterprises that need it.

Model Marketplace
Bigger than people are giving it credit for
Enterprises can buy open-weight inference from Baseten through the Responses API and Codex using their OpenAI spend. It insulates OpenAI's moat from open-source disruption and makes big commitments safer for anyone running a multi-model strategy.

What Dev Day Says About AI
Nothing blisteringly new, just confirmation of the trends
Models good enough and cheap enough to do everything (6.1 Sol, Decisions API), persistent agents that actually do it (Dots), bringing everything in and going everywhere (plugin extensions, Sign in with ChatGPT), and native multiplayer AI (Space), even if that last one is still nascent.