Search
On February 12, 2026, OpenAI launched its inaugural production-ready AI coding model, GPT-5.3-Codex-Spark, which operates on non-Nvidia hardware. Powered by Cerebras' Wafer Scale Engine 3 chip, the model delivers inference speeds surpassing 1000 tokens per second, marking a roughly 15-fold increase compared to its forerunner. Emphasizing speed, the model initially supports text processing exclusively and boasts a context window of 128,000 tokens. It surpasses older iterations in programming benchmark tests, achieving superior results in a shorter timeframe. Presently, it is accessible as a research preview for ChatGPT Pro subscribers, with API access being incrementally introduced. This launch signifies the inaugural milestone in the partnership between OpenAI and Cerebras, aimed at broadening the range of computing power providers and diminishing dependence on Nvidia.
-
C114 Communication Network -
Communication Home
