# The Most Important New AI Tools from OpenAI DevDay *The AI Daily Brief — Wednesday, 2026-09-30 · https://aidailybrief.ai/e/2026-09-30* **DevDay didn't reveal a new direction — it confirmed the one we're already on.** OpenAI's biggest DevDay ever wasn't about something blisteringly new. Sol 6.1 and the Decisions API deepen the move to models cheap and good enough to be used for everything. Dots turns that into persistent, proactive agents that actually do everything. Plugin extensions and sign in with ChatGPT bring everything in and go everywhere. And Space confirms that the next generation of AI won't just be single-player — it's going multiplayer, natively. These are the trends that will reshape how we all use AI in the months to come. --- ## By the numbers - **20+** — Launches and announcements at a single DevDay - **40,000** — Apps Dots can work across on its own cloud computer - **75.2%** — GPT-6.1 Sol's DeepSUI score on high — edging Astra's best - **13%** — 6.1 Sol's cost relative to Astra (a third the cost of 6 Sol) - **1.2B** — Weekly ChatGPT users now open to plugin-extension developers - **$500** — New subscription tier with 25× the usage of Plus - **8×** — Ultra Fast Mode token speedup in Codex - **50%** — Effective usage reduction on the reopened $200 Pro tier ## Main episode ### Astra pulled OpenAI's 2027 roadmap into 2026 `[00:00]` The OpenAI team says the latest generation of models like Astra increased their internal development speed so significantly that things they expected to ship in 2027 arrived as part of DevDay — more than 20 launches and announcements in all. *For: Exec* Link: https://aidailybrief.ai/e/2026-09-30#astra-pulled-2027-forward ### Dots: OpenAI's answer to Muse and GrokBot `[02:00]` Sam Altman pitched Dots as "remarkably capable, always-on agents built to handle really anything you can think of." Users create a persistent agent, talk to it through a text-message-style interface, and set it loose on tasks using its own cloud computer with support for 40,000 apps — plus voice calls, Teams, and Slack with context carried across services and sessions. *For: Ops, Product* Link: https://aidailybrief.ai/e/2026-09-30#dots-personal-agent ### Dots skips the free tier that made Muse a hit `[03:00]` For now it's one dot per user, available only to Pro, Business, and Enterprise customers — a deliberate limit on competing with the completely free Muse, though CFO Sarah Friar says the vision is the whole consumer base. The selling point: GPT-6 Astra under the hood, the first time a truly frontier model pilots a first-party personal agent. *For: Product, Finance* Link: https://aidailybrief.ai/e/2026-09-30#dots-skips-the-free-tier ### Early Dots verdict: flashes of excellence, plenty of bugs `[04:00]` Users reported lost work sessions and no way to connect multiple computers for full context, while others had it nail autonomous expense work that Grokbot botched. Every's vibe check lands at TBD — truly excellent at moments, extremely frustrating at others. The live-demo hiccups matter less than the chattering classes think; demos going wrong is how you know they're actually live. *For: Ops* Link: https://aidailybrief.ai/e/2026-09-30#dots-teething-problems ### Dots is more coworker than companion `[04:00]` Early users describe Dots as more work-focused than Grokbot or Muse — less personal assistant, more Codex orchestrator and Slack collaborator. Slack is where it reportedly shines: your whole team can message the bot directly and have it take actions. It's also fueling a debate over whether this should have been a separate app from ChatGPT entirely. *For: Eng, Ops* Link: https://aidailybrief.ai/e/2026-09-30#dots-codex-orchestrator ### Form factor is converging — utility isn't `[05:00]` The personal-agent shape has been inevitable since Open Claw, but as Nate B. Jones argues, each player's actual utility is distinct: Muse is good at phone calls and practical email, Instinct at travel, Dots at AI context as a work surface. The trillion-dollar question is whether each picked the right utility to get right. *For: Product* Link: https://aidailybrief.ai/e/2026-09-30#form-factor-vs-utility ### The Decisions API is OpenAI's answer to Jev `[06:00]` A version of Luna that mimics Jev's rapid classification and judgment abilities: define questions and possible answers, get fast outputs for classification, routing, or choosing an agent's next action — 10× faster decision-making than the Responses API, per OpenAI. One real differentiator: visual inputs, which Jev can't handle natively. For now it's preview-only with a limited group. *For: Eng, Product* Link: https://aidailybrief.ai/e/2026-09-30#decisions-api ### Judgment models are becoming a fundamental primitive `[07:00]` The Decisions API is a recognition that Jev represents more than a new model. LLMs were being rammed like square pegs into round holes for classification and judgment use cases — once you see it, a judgment model sitting alongside your generative model seems obvious in retrospect. Expect every frontier lab to ship some version of this very soon. *For: Eng* Link: https://aidailybrief.ai/e/2026-09-30#judgment-models-are-a-primitive ### Space is OpenAI's AI-native answer to Google Drive `[08:00]` A shared document workspace where human teams and agents collaborate on spreadsheets, decks, and docs — and where workplace automations live. Dots work on Space documents natively, and teams can schedule tasks to produce deliverables. Microsoft 365, Drive, and Notion have all felt increasingly like a bad fit for agentic work; a truly AI-native productivity suite from OpenAI was only a matter of time. *For: Ops, Product* Link: https://aidailybrief.ai/e/2026-09-30#openai-space ### The power users' instant verdict: Dots overhyped, Space underhyped `[09:00]` While nearly every other announcement drew mixed reviews, Space won power users immediately — every company wants an AI-native collaboration workspace, as Claire Vo put it. Dan Shipper's key point is speed: when an agent edits Google Docs through a browser, changes crawl because Docs wasn't built for agents; native documents have no lag, and tagging your dot inside the document feels like actual collaboration. *For: Product* Link: https://aidailybrief.ai/e/2026-09-30#space-underhyped ### GPT-6.1 Sol: near-Astra intelligence at a fifth the price `[10:00]` Just a week after GPT-6 Sol, coding benchmarks are up significantly across all effort levels — 6.1 Sol on medium outperforms 6 Sol on max, and its 75.2% DeepSUI score on high actually edged past Astra's best, supporting OpenAI's near-Astra claim at a much cheaper price. *For: Eng, Finance* Link: https://aidailybrief.ai/e/2026-09-30#gpt-61-sol ### Max effort settings are starting to hurt performance `[10:00]` 6.1 Sol's extra-high and max settings actually degraded its DeepSUI results versus high. First seen with Opus 5, the pattern suggests top effort settings are forcing models to overthink and second-guess correct responses to their own detriment. *For: Eng* Link: https://aidailybrief.ai/e/2026-09-30#max-effort-overthinking ### A quiet breakthrough in computer-use efficiency `[11:00]` On OSWorld, which tests long-horizon computer use, 6.1 Sol scored 71.4% — beating 6 Sol's 64.4% and nearing Astra's 73.5% — at a third of 6 Sol's cost and 13% of Astra's. And raising the effort level barely changed the cost, suggesting real efficiency breakthroughs under the hood. *For: Eng, Finance* Link: https://aidailybrief.ai/e/2026-09-30#computer-use-efficiency ### Fifth place on intelligence, first on cost-efficiency `[11:00]` Artificial Analysis scored 6.1 Sol at 52 — a point shy of Astra and behind Opus, Sonnet, and Fable — but found its benchmark run cost a quarter of Astra's and 31% less than 6 Sol's. Below max settings, it approaches the cost territory of GLM 5.3 Flash. OpenAI calls it the most cost-efficient model for its performance available today; early hands-on reports are split. *For: Finance* Link: https://aidailybrief.ai/e/2026-09-30#pareto-frontier-pricing ### For anything regarding safety and alignment, there's a trade-off. `[15:00]` *— Saachi Jain, OpenAI's head of safety systems* The Wall Street Journal reports OpenAI scrapped plans to release GPT-6.1 Astra over safety concerns: they tried to turn down the tenaciousness behind recent incidents but couldn't find a happy medium. The model "didn't quite meet the bar in terms of staying within scope and authorization," per Jain — suggesting technical breakthroughs are needed before the next frontier level ships at all. *For: Exec* Link: https://aidailybrief.ai/e/2026-09-30#safety-alignment-tradeoff ### Plugin extensions: the ChatGPT app store play, again `[16:00]` Developers can now build full native apps that ship right inside ChatGPT, surfaced to 1.2 billion weekly users in conversation. OpenAI has been trying to get this balance right for over a year across integrations, plugins, and MCP. The skeptics' question: if you bring the app and OpenAI brings the intelligence — and possibly the user traces — are you just teaching your partner how to do your job? *For: Product* Link: https://aidailybrief.ai/e/2026-09-30#plugin-extensions-app-store ### Sign in with ChatGPT is a pricing story, not an auth story `[17:00]` Users carry their intelligence subscription into third-party apps, so developers don't have to justify a separate subscription or price everything on top of API costs. As Jackie Luo argues, this is OpenAI extending favored pricing to third parties and bundling services into the subscription — individuals stop double-paying for tokens, apps charge for the app layer alone, and Anthropic will likely need to follow. *For: Product, Finance* Link: https://aidailybrief.ai/e/2026-09-30#sign-in-with-chatgpt ### Ultra Fast Mode and a new $500 tier for power users `[18:00]` Ultra Fast Mode delivers 8× faster token production in Codex and 6× in the app — for now Astra and Codex only in ChatGPT Work, with 6.1 Sol support coming. A new $500/month tier offers 25× the usage of Plus and is currently the only tier with Ultra Fast Mode access. *For: Eng, Finance* Link: https://aidailybrief.ai/e/2026-09-30#ultra-fast-500-tier ### Compute constraints are real, present, and permanent `[19:00]` The $200 Pro tier reopened with usage calculations netting out to a 50% reduction in API-cost terms — the best of a bad set of choices, per OpenAI, to avoid five-hour usage limits. Frontier models are running up against the walls of available compute, and new supply isn't coming online anywhere near the speed demand is growing. The upside: a big efficiency push that should net better, cheaper experiences — with bumps along the way. *For: Finance, Exec* Link: https://aidailybrief.ai/e/2026-09-30#compute-constraints-permanent ### Codex goes always-on: close the laptop, work continues `[20:00]` Codex now has a dedicated cloud environment so work persists after you shut your machine — inevitable once thumbs-jammed-into-laptops became a San Francisco meme. Starting with GrokBot, the whole industry has moved toward cloud instances as a necessary part of agentic products: AI shifting from software you engage with to a persistent, always-on application. *For: Eng* Link: https://aidailybrief.ai/e/2026-09-30#codex-always-on-cloud ### Private Intelligence: zero data retention, even at inference `[21:00]` With growing concerns that OpenAI and Anthropic are skimming data from their enterprise customers — essentially the main text of every Microsoft communication these days — a stronger, clearer end-to-end data privacy guarantee is a big deal for the enterprises that need it. *For: Legal, Exec* Link: https://aidailybrief.ai/e/2026-09-30#private-intelligence ### The sleeper hit: an open-weight model marketplace `[21:00]` Enterprises can now spend OpenAI credits on open-weight model inference from Base10 through the Responses API and Codex. This has big implications for OpenAI's moat — it protects them from open-source disruption while letting customers make large spending commitments knowing they can run a multi-model strategy. It's bigger than people are giving it credit for at first blush. *For: Finance, Exec* Link: https://aidailybrief.ai/e/2026-09-30#open-weight-marketplace ### DevDay confirmed the trends rather than breaking new ground `[22:00]` Sol 6.1 and the Decisions API deepen the era of models good enough and cheap enough to do everything without rationing. Dots turns that into persistent, proactive agents. Plugin extensions and sign in with ChatGPT mean bringing everything in and going everywhere. And Space confirms the next generation of AI will be team and multiplayer mode in a native way — even if that remains nascent so far. *For: Exec* Link: https://aidailybrief.ai/e/2026-09-30#devday-confirmed-the-trends *Today's sponsors: KPMG, Blitzy, Robots and Pencils, Hyperagent — offers at https://aidailybrief.ai/sponsors* --- Transcript: https://aidailybrief.ai/e/2026-09-30/transcript.md Listen: https://pod.link/1680633614 · Ad-free: https://patreon.com/aidailybrief © 2026 The AI Daily Brief — Until next time, peace ✌