OpenAI is refining its inference software with its own models on NVIDIA GPUs, and NVIDIA notes ongoing optimization can speed responses while boosting infrastructure productivity, with a programmable platform that enables reuse of the same infrastructure across training, inference, and reinforcement learning to meet demand.

NVIDIA AI infrastructure delivers higher throughput and lower latency, helping OpenAI serve more useful model outputs, and the platform’s reuse across training, inference, and reinforcement learning improves resource utilization and adaptability.

The GPT-6 family has expanded to Copilot, with GPT-6 Sol and GPT-6 Luna available to Copilot users and billed on usage, rolled out across editors and surfaces like Visual Studio Code, JetBrains, Xcode, Copilot CLI, github.com, and GitHub Mobile.

Faster responses are particularly valuable in time-sensitive workflows where agents write code, use tools, and iterate on results.

Ultrafast delivers up to eight times faster token generation than Astra Standard, enabling quicker edit-test-debug cycles for developers and more responsive interactive applications.

The Ultrafast model runs on NVIDIA Blackwell GPUs and achieves up to 8x faster token generation due to inference optimizations tailored to Blackwell’s architecture.

Developers can access GPT-6 Astra Ultrafast via the OpenAI API today, with the Ultrafast guide outlining access, pricing, and implementation details.

OpenAI’s GPT-6 Astra Ultrafast, running on NVIDIA Blackwell GPUs, is available in the OpenAI API and to eligible ChatGPT Work and Codex users.

The API access guide covers how to get Ultrafast, but does not disclose prices, token limits, context windows, benchmarks, or regional availability.

GitHub Copilot Business and Enterprise admins manage model access via policy, with default enablement for new models and the option to adjust defaults or disable models; several older models are slated for deprecation on October 19 across Chat, inline edits, ask, agent, and completions.

NVIDIA emphasizes practical benefits of faster generation, including shorter edit-test-debug cycles for coding agents, reduced wait times between tool calls, and more responsive interactive applications.

The OpenAI–NVIDIA collaboration centers on inference optimizations and the use of Astra to generate high-performance kernels on Blackwell and Rubin GPUs, supported by NVIDIA’s tooling and documentation.