At its DevDay developer conference on September 29, 2026, OpenAI shifted its focus away from purely open-ended conversational models toward deterministic backend infrastructure. The company is directly addressing demands from institutional developers who deploy machine learning in heavily regulated sectors like finance. In these operational settings, unpredictable free-text outputs and latency spikes previously posed significant integration obstacles.
The centerpiece of this update is the new Decisions API, launched in a limited preview and powered by a specialized model called Luna. Rather than generating descriptive text, the interface maps multimodal inputs, such as text, scans, and documents, directly onto a predefined, fixed catalog of outcomes in real time. This architecture completely bypasses conventional prompt parsing and drastically reduces the hallucination risks common in classification tasks.
For financial institutions and fintech platforms, this mechanism unlocks direct automation across high-stakes workflows. Latency-critical routines, often referred to as system-one decisions, can run directly through the API for identity verification (KYC), fraud detection, anti-money laundering reviews, and trade order routing. Engineers receive deterministic status responses without statistical text anomalies or parsing errors.
OpenAI also moved its Agents API with integrated computer-use capabilities into public beta. This platform enables autonomous agents to interact with web browsers and legacy desktop user interfaces in the same manner as human operators, executing multistep tasks across disparate business software. To minimize local security risks and latency, OpenAI runs these agent sessions in isolated, company-hosted cloud environments.
To serve algorithmic traders and high-frequency data pipelines, OpenAI unveiled GPT-6.1 Sol alongside an execution tier dubbed Ultrafast. This premium inference tier delivers up to eight times faster token throughput than standard enterprise deployments. It is explicitly tailored for programmatic execution setups where transaction speed takes precedence over marginal API cost.
Alongside these technical specifications, OpenAI announced that ChatGPT has expanded to 1.2 billion weekly active users. Together, the introduction of deterministic classifications and computer-use automation signals OpenAI's transition from consumer-facing productivity software into mission-critical infrastructure for enterprise operations.