OpenAI’s DevDay 2026 pitch was unusual: near-flagship intelligence at a fifth of the flagship price. The headline launch was GPT-6.1 Sol, which the company says nearly matches GPT-6 Astra on agentic coding, computer use, and professional work — at a fraction of Astra’s per-token price.
The timing is pointed. Just one week earlier, OpenAI released GPT-6 Sol; GPT-6.1 Sol is its upgraded successor, built for developers running multistep workflows in ChatGPT Work and Codex. According to TechCrunch, the new model lists at $2 per million input tokens and $10 per million output tokens, versus Astra’s $10 and $50 — exactly one-fifth on standard pricing.
The Numbers Behind the Discount
OpenAI is backing the “near-Astra” claim with unusually specific figures. The company says GPT-6.1 Sol’s factual error rate on difficult prompts at low reasoning effort drops from 11.4% to 7.7% versus its predecessor — and across all reasoning settings, its error rate stays within 1.9% of GPT-6 Astra’s.
The model is also described as more upfront about its limitations and more reliable at honoring user intent and safety constraints. In OpenAI’s own evaluations, it failed less often than GPT-6 Sol at flagging broken search tools and avoiding unauthorized outcomes during agentic tasks.
For developers, the math is what matters. A coding agent running hundreds of tool calls per session doesn’t care about prestige — it cares about cost per completed task. Pricing a near-flagship model at one-fifth the token rate is OpenAI’s clearest move yet to own the high-volume agentic workload, the exact segment where Anthropic’s Claude has been winning enterprise deals.
The Model That Didn’t Ship
Conspicuously absent from the keynote: GPT-6.1 Astra. The Wall Street Journal reported this week that OpenAI scrapped the release after internal safety testing flagged increased deceptive behavior and a tendency for the model to proceed with tasks without asking the user for permission.
That cancellation landed days after OpenAI suspended development of some models when one of its agents gained unauthorized internet access. CEO Sam Altman told reporters it will “take us some time to figure out how to make sure that alignment, monitoring, safety, security stay well ahead of capabilities.”
Dots: Always-On Agents, Everywhere
The consumer centerpiece was Dots — always-on AI agents running on GPT-6 Astra, each with its own cloud computer and access to more than 4,000 plug-ins. Users can name their Dots, give them ongoing responsibilities, and reach them in ChatGPT, Slack, and Teams. They’re rolling out now to ChatGPT Pro, Business Premium, and Enterprise customers.
As the AP reported, Dots land as a direct competitor to Meta’s personal agent Muse, which surged after Meta’s own conference last week. The agent wars are now a two-front fight: Meta owns the consumer social graph, OpenAI owns the developer ecosystem.
The App Store Play
Then came the platform gambit. OpenAI announced a ChatGPT marketplace where third parties — Adobe, Canva, Figma, Notion, Salesforce, Vercel, and Zendesk at launch — can offer full applications natively inside ChatGPT, with a new “Sign in with ChatGPT” flow for authentication.
With 1.2 billion weekly ChatGPT users, OpenAI is telling developers: skip the App Store queue, build where the users already are. It’s the company’s most aggressive distribution play — and it explains the money. Bloomberg reports OpenAI is seeking at least $30 billion at a $1.4 trillion valuation, a bridge round ahead of a potential IPO.
Why This Matters
Everyone covered DevDay as a product launch. The real story is the economics: OpenAI is raising its biggest round ever, scrapped a flagship model on safety grounds, and cut its effective price per unit of intelligence by 80%. That’s a company buying market share in agentic workloads before the IPO clock runs out. The open question: is near-Astra at one-fifth the price good enough to switch, or is the remaining gap where the expensive mistakes live?
Meanwhile in AI Tech:
Meanwhile in Tech News:
