Capability change
- Before: Computer interaction was not established here as a generally available, priced model feature builders could select through public APIs and products.
- Enabling change: Google's Gemini API computer-use release and OpenAI's GPT-6.1 Terra computer-use offering.
- After: Builders can use generally available API computer use or choose a fast, cheaper computer-use model in OpenAI's products and API.
Human consequence: People building agents can procure computer interaction as a model feature rather than treating it as a bespoke integration.
Bottleneck: The items do not specify reliability or limits; computer-use failures can still result from page-loading races.
Buildability: B4. Verification: source-inspected. Related shift: coding-agent-general-workbench.
Two items support a concrete shift: computer use is described as generally available in the Gemini API and as a priced model feature from OpenAI. Research also identifies page-loading races as a persistent failure source, so this is availability, not proof of dependable automation.
Build around the released model capability and check actions and page readiness, not only final task success. Watch whether reliability improves and whether the named availability extends beyond these offerings.
Published automatically by the New Runtime weekly knowledge desk from the week's evidence.
Supporting evidence
- Computer use has become a standard, priced model feature you can build on (signal, 2026-11-06)
- Agent reliability now needs checks on each action, not just the final result (signal, 2026-11-03)
Provenance
Auto-created by the New Runtime weekly knowledge desk on 2026-11-09 from evidence reviewed in the week of 2026-11-02. It is published through the live overlay and joins the hand-edited catalog at the next release.