OpenAI ships always-on agents to beat Meta and xAI to market

OpenAI used its DevDay keynote in San Francisco to stake a claim on the always-on agent market, launching Dots a day after quietly pulling GPT-6.1 Astra for failing internal safety checks. The timing puts pressure on Meta and xAI, both of which beat OpenAI to always-on agents this year, and forces the question of whether OpenAI's answer arrives strong enough to close the gap.

Dots run on GPT-6 Astra, each getting a dedicated cloud computer and browser to work toward user-set goals continuously, connecting to over 4,000 apps via plugins and reachable through ChatGPT's apps, Slack, Microsoft Teams and voice. Users can set approval rules per action; password changes always require sign-off. The first dot is free with ChatGPT Pro ($200) and Business Premium plans, with enterprise and education versions in admin-approval beta. As OpenAI put it, "Your first dot is included in your Pro or Business Premium plan at no extra cost." Dots go head-to-head with Meta's Muse, launched 8 September, and SpaceXAI's Grok Bot, in beta since 11 August.

Alongside the agents, OpenAI shipped GPT-6.1 Sol, a week after GPT-6 Sol, pricing it at $2 per million input tokens, $0.10 cached and $10 output — roughly a fifth of GPT-6 Astra's rates. OpenAI says it nearly matches Astra on agentic coding, computer use and professional work. Third-party numbers from Artificial Analysis put its Intelligence Index at 52 against 58 for Anthropic's Claude Opus 5.5, but at $0.72 per task versus $5.98 for Opus 5.5 — an 8x cost gap for a 6-point capability shortfall. The model is live in the API and in ChatGPT Work and Codex.

OpenAI also introduced Ultrafast, a premium speed tier running up to 8x faster in Codex and 6x in the API at up to 6x the price ($60 input/$300 output per million tokens for GPT-6 Astra), and a new Pro 500 plan at $500 a month offering 25x Plus usage with Ultrafast bundled in. That comes alongside a quiet repricing of the existing $200 Pro tier, whose usage allowance drops from 20x to 10x Plus for new subscribers, according to Decrypt. The $100 tier holds at 5x. More than 20 other products landed too, including ChatGPT Space, Pages, plugin extensions, Sign in with ChatGPT with 16 launch partners including Notion and Vercel, and an Agents API with computer use.

OpenAI is compensating for a stalled flagship by pricing aggressively at the mid-tier and racing agents to market before rivals consolidate their lead. GPT-6.1 Sol's cost-per-task advantage over Claude Opus 5.5 makes it the pragmatic choice for high-volume agentic workloads, even if Anthropic keeps the raw capability edge. The Pro tier downgrade, offset by a pricier Pro 500 option, signals OpenAI is segmenting power users into a higher-margin bracket rather than absorbing Ultrafast's costs across the base. The bigger exposure is Astra: a cancelled release over safety standards, one day before this launch, raises the question of whether OpenAI's next flagship model is further behind than its Sol pricing strategy suggests.