The headline of the day is the model and pricing stack. GPT-6.1 Sol ships at $2 per million input and $10 per million output tokens — the same list as GPT-6 Sol a week earlier — but cached input drops to $0.10 per million, 95% below standard pricing and half of GPT-6 Sol’s cache tier. OpenAI calls it “near-Astra intelligence at one-fifth of Astra’s price”: Artificial Analysis puts Sol at 52 on its Intelligence Index against Astra’s 53, at roughly 22% of Astra’s cost per task. At the top end, a new Ultrafast tier runs GPT-6 Astra at up to 300 tokens per second at six times the standard rate ($60/$300 per million), and ChatGPT’s new Pro 500 plan — $500 a month, 25 times the Plus allowance, Ultrafast included — reopened the $200 tier on stage.

The platform turn is the bigger story for enterprises. Dots are always-on agents powered by GPT-6 Astra, each with its own cloud computer, a browser and more than 4,000 supported apps, living inside ChatGPT, Slack and Teams; the first dot is included in Pro and Business Premium, and conversations with it do not count against usage limits. OpenAI also shipped an Agents API with hosted computer use, Codex in the cloud with repo-wide security scanning, Sign in with ChatGPT (a user’s own plan pays for usage inside third-party apps, starting with 16 partners), and the OpenAI Marketplace, where enterprise commitments can be spent on partner software from Adobe, Figma, Harvey, CrowdStrike and Baseten. Most consequential for Indian buyers: Bedrock Managed Agents, powered by OpenAI, puts the lab’s agents onto AWS through Amazon Bedrock.

The safety shadow is not cosmetic. A day before the keynote, WSJ and CNBC reported that OpenAI shelved GPT-6.1 Astra — the next-generation flagship planned for an October debut — after internal safety testing found it did not adequately meet the lab’s standards; head of safety systems Saachi Jain cited an “extremely high bar.” Sam Altman told reporters OpenAI will not go public until it can make “confident safety claims,” while warning that waiting too long would be “bad for the world.” CFO Sarah Friar put the commercial counterweight on the record: about 70% quarter-on-quarter growth in Q3, an enterprise business that has doubled since July, an annualised run rate approaching $70 billion, and early-stage talks with investors on a new round. Protesters outside Fort Mason pressed the same week’s themes: the ICE contract and data-centre energy.

The most-teased product of the week did not ship. GPT-6 Cyber, the fourth cybersecurity model of 2026 that Fortune reported would preview at DevDay alongside a gated security-deployment product, did not appear in launch-night recaps — the dev.to pricing table, The Verge and CNBC all omit it — leaving the preview window open for “coming weeks.” For Indian BFSI, telecom and GCC security teams, the who-gets-access-and-at-what-price question for gated cyber capability stays the open variable, even as Grok 4.7 reached Amazon Bedrock on September 28 for the same AWS bill and guardrail framework.

For Indian enterprises, two-speed pricing is the planning variable: workhorse deflation (Sol cache at −95%) versus a premium agentic speed tier ($500 Pro, $60/$300 Ultrafast), with hosted agents on AWS and user-paid tokens changing procurement and cost-routing. Markets, for context only: DevDay was the same-day catalyst Indian IT did not get — Nifty IT is down roughly 11% in September with Infosys, Wipro and Tata Elxsi at 52-week lows, and BSE replaces Wipro in the Nifty 50 effective today. Sensex closed Tuesday at 72,529.07 (−242.65), Nifty at 22,716.20, the rupee near a two-month low at 96.13, and Brent eased to about $105 as US-Iran indirect talks revived.