OpenAI ships Agents API public beta, pauses ChatGPT Pro signups on Astra demand

OpenAI released its Agents API into public beta as a managed service built on the Codex harness, offering orchestration, long-running sessions, and context management, and deployable across OpenAI's own sandboxes, customers' sandboxes, or partner environments. The launch is aimed squarely at reducing the 'glue code' developers write to string together agentic workflows, with native Model Context Protocol (MCP) support.

Developer reception was strong: the release hit 248 points on Hacker News, with practitioners calling it 'stupidly easy to build real AI agents' (Min Choi) and praising native MCP support, the absence of an extra harness fee, and orchestration that 'eliminates an absurd amount of glue code.' The positive signal matters because agent frameworks have proliferated and OpenAI is betting a first-party managed offering can consolidate the fragmented tooling market.

Simultaneously — but as a separate operational move — OpenAI paused new $200/month ChatGPT Pro signups, citing 'unprecedented' demand following the September 3 launch of GPT-6 Astra. The pause underscores how capacity-constrained frontier serving has become, echoing AWS's decision to bring Astra to Bedrock to spread the load across governed enterprise infrastructure.

What to watch: whether the Agents API's managed model wins developers away from open frameworks, how pricing settles after beta, and how long the Pro signup pause lasts as OpenAI scales compute for Astra demand.