On July 22, 2026, OpenAI announced Presence — an end-to-end managed platform for deploying AI voice and chat agents inside companies. It isn't a new model. It isn't another API. It's a product that includes operating policies, guardrails, connections to internal systems, pre-launch simulations, evaluations and a continuous-improvement loop powered by Codex.
The move marks a fundamental expansion of OpenAI's business: from capability provider to service operator. The distinction has deep implications for the enterprise software market.
What Presence has that the API doesn't
The core difference is post-launch governance. With the API, the client company builds, deploys and manages — including keeping behavior consistent when internal policies change, validating changes before rolling them out, and improving continuously without compromising production systems. Presence does all of that.
The technical components include standard operating policies that define what the agent can and can't do; configurable guardrails to limit behavior outside the defined scope; approved actions such as identity verification, data access and processing refunds; pre-launch simulations to validate changes before they reach production; and the continuous-improvement loop powered by Codex, which tunes the agent based on real performance.
The technology behind Presence
Presence's voice engine is GPT-realtime, OpenAI's model for low-latency synchronous conversations. With the Presence launch, GPT-realtime audio prices dropped 20%: roughly $32 per million input audio tokens and $64 per million output tokens.
The service requires certified engineers from the OpenAI Deployment Company — OpenAI's professional services division, valued at $10 billion in its latest round. There's no self-service version. Contracts are annual with guaranteed minimum consumption — the classic enterprise sales model, not self-service SaaS.
First customers and the market signal
BBVA Mexico uses Presence to improve personalized interactions with banking customers — a high-volume use case where the agent's consistency directly affects the customer experience. SoftBank Corp Japan is deploying it for Japanese-language agents that can hold natural conversations, a market where language quality is critical for adoption.
The pattern from these first two customers is clear: large companies in markets where voice agents have cultural and regulatory traction — banking and telecoms in markets with demanding customer-service standards.
The strategic move and its implications
Presence competes directly with Salesforce, ServiceNow and Zendesk in customer experience management. What makes this politically complex is that Salesforce, ServiceNow and Zendesk are current OpenAI partners — companies that have built products on OpenAI's API. Presence doesn't take the API away from them; it takes away the end customer.
The tension in the ecosystem is real. Google responded by investing $750 million in its ecosystem of consultants and integrators. Anthropic launched its own platform, Ode, with a similar concept — a sign that the convergence toward managed services is a structural trend, not an isolated bet by OpenAI.
The open question is whether Presence will eventually offer a self-service version for mid-sized companies, or remain what it is today: AI-powered luxury consulting for large accounts with real enterprise budgets.
- No self-service version — requires certified engineers
- Annual contracts with guaranteed minimum consumption
- GPT-realtime: audio 20% cheaper with the launch
- Initial customers: BBVA Mexico and SoftBank Japan
- Competes directly with Salesforce, Zendesk and ServiceNow
Frequently asked questions
What is OpenAI Presence and how does it differ from the API?
Presence is an end-to-end managed platform for deploying voice and chat agents in companies. Unlike the API (where you build and manage), Presence includes governance, simulations, continuous improvement and support from OpenAI engineers from design through operation.
Can any company access Presence?
No. Presence requires certified OpenAI engineers or certified integrators. There's no self-service option. Contracts follow an enterprise structure with annual terms and guaranteed minimum consumption.
What technology does Presence run on?
It's built on GPT-realtime for synchronous voice conversations and Codex for the continuous-improvement loop. GPT-realtime audio prices dropped 20% with the launch.