OpenAI used DevDay 2026 to launch Dots, always-on agents powered by GPT-6 Astra that run on their own cloud computers and connect to more than 4,000 apps plus Slack and Teams. The same event brought GPT-6.1 Sol, pitched as near-Astra intelligence for a fifth of the price, alongside an Ultrafast generation mode, a Decisions API, ChatGPT Spaces and Pages, and a B2B marketplace. For companies buying AI, the message is lower cost per task paired with agents that keep working with a laptop closed.
Dots, Sol pricing and platform updates
Each Dot operates under user-set boundaries covering what it can do alone, what needs approval, and what it must never do. Connecting a personal machine is optional, and the product ships to Pro, Business Premium and Enterprise plans. The primary Dot's direct work does not draw on plan usage, while Codex tasks it spawns do. Developers can hand off bug triage, failing builds and pull requests through Codex, and early testers cited proactive cases such as negotiating with customer service to cut about $500 per year in charges.
GPT-6.1 Sol is priced at $2 for input and $10 for output per million tokens, with cached input at $0.10, a 95% cache discount. OpenAI claims it ties Astra on DeepSWE, beats Opus 5.5 on AutomationBench at one third of the cost, and lands 2.1 points short of Astra on OSWorld 2.0 at about one seventh of the cost. The company also reports about 32% fewer factual errors on hard prompts versus GPT-6 Sol and stronger alignment evaluations. Independent checks add texture: Artificial Analysis places Sol one point below Astra on its Intelligence Index at $0.72 versus $3.26 per task, with hallucination rate down from 60% to 54%.
Ultrafast mode offers up to 8x faster generation at 300 tokens per second in Codex and 6x in the API, priced at 6x, or $60/$300 per million tokens for Astra. The Decisions API provides near-instant multiple-choice classification and routing on GPT-6 Luna over text and images, described by observers as a competitor to routing approaches discussed in the Jev podcast. Codex also gained cloud environments that continue running, a refreshed CLI with worktrees and /agents, and Security Cloud. Enterprises can apply OpenAI commits to open models via Baseten, a move framed as a bid to own more of the enterprise AI budget.
What this means for AI adoption in business
For operating teams, Dots plus Spaces and Pages point to shared human-agent workspaces where an agent monitors apps, prepares work, and requests approval only at defined thresholds. A small firm could assign one Dot to vendor billing, inbox triage, or build failures without provisioning infrastructure, since each agent runs on its own cloud computer. A large organization can standardize boundaries across departments and route repetitive approvals through the Decisions API, while keeping Codex execution metered separately. The practical shift is from chat sessions to delegated background work with audit points.
The cost structure changes model selection. At $2/$10 with $0.10 cached input, Sol becomes the default for high-volume coding, classification, and automation, while Astra remains for peak difficulty. One planted-bug test cited 44 bugs found for $6.56 on 6.1 Sol versus 45 for $33 on Astra and 41.7 for $58.53 on Opus 5.5, and a vision test showed 81.6 mAP@50 versus 83.6 for Astra at 78% lower cost. At the same time, plans were re-tiered to Plus 1x, Pro 100 5x, Pro 200 10x, plus a new Pro 500 at 25x, a change that roughly halves the old Pro 200 value and drew backlash. Buyers should model total spend including Ultrafast premiums and Codex-spawned usage.
The marker to watch is sustained Sol performance inside company harnesses rather than launch benchmarks. Artificial Analysis and independent Codex-harness runs diverged, with harness sensitivity under debate, and Sol uses 10-30% more output tokens than GPT-6 Sol. Track defect-find rates, AutomationBench-style task completion, and cost per resolved ticket over the next quarter. If Sol holds near-Astra results at one fifth of the price in production logs, routing most agentic workloads to it will become standard practice.
