TL;DR
OpenAI's DevDay 2026 moved the company from selling model calls to selling persistent agent infrastructure, launching computer use in the Agents API, a low-cost Decisions API, cloud Codex environments, and security scanning in a single sweep.
What happened
- DevDay 2026 took place in San Francisco, with OpenAI announcing a dense stack of developer and agent platform updates on September 30, 2026.
- Computer Use is now available in the Agents API (public beta since September 11), letting any developer-built agent operate software through graphical interfaces, not just Codex and ChatGPT users who got it in April and May.
- GPT-6.1 Sol launched as a coding and professional-task model that approaches GPT-6 Astra quality on several benchmarks at one-fifth the token price, with cached input at $0.10 per million tokens.
- Decisions API (limited preview) uses a version of GPT-6 Luna for fast, single-choice tasks like classification and request routing, putting OpenAI directly into the emerging "System One" fast-decision model category.
- Codex Security Cloud can scan entire GitHub repositories on demand or on a schedule, investigate findings, remove duplicates, and prepare fixes automatically, with Daybreak Blue cybersecurity model access bundled in.
Why it matters
- Computer use in the API is the biggest unlock: any developer can now build agents that operate real software, not just call tools, collapsing the gap between AI assistants and autonomous workers.
- GPT-6.1 Sol's pricing (one-fifth of Astra) creates a credible high-quality, low-cost option that undercuts the cost argument for rivals at the coding and professional-task tier.
- The Decisions API signals OpenAI is chasing the fast-routing model niche that TypeSafe AI's Jev model just validated, pairing a strong orchestrator with a cheap, fast decision layer for agentic pipelines.
- Bedrock Managed Agents (limited preview with AWS) means OpenAI agents can now run entirely on AWS infrastructure, deepening the Amazon partnership and giving enterprise buyers a familiar compliance envelope.
- Dots and ChatGPT Space (persistent agents and shared team workspaces) extend the platform toward continuous, collaborative AI work rather than one-shot queries, raising the switching cost for enterprise customers.
What to watch next
- Decisions API general availability: if Luna-based routing proves reliable at scale, it could become the default plumbing layer inside multi-agent pipelines across the industry.
- Adoption of Codex Security Cloud: security teams are a new buyer persona for OpenAI; enterprise contract sizes and compliance certifications will determine whether this is a real wedge or a feature.
- Competitor responses on computer use pricing: Anthropic and Google both offer computer use; watch whether GPT-6.1 Sol's cost advantage forces a price response or accelerates developer consolidation around OpenAI's agent stack.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.