Krux
February 19, 2026
OpenAI Drops Lightning-Fast Coding AI Powered By Cerebras
Published: February 19, 2026 at 1:03 AM
Updated: February 19, 2026 at 1:03 AM
100-word summary
OpenAI just launched GPT-5.3-Codex-Spark, a blazingly fast coding model that churns out 1000+ tokens per second thanks to Cerebras' massive wafer-scale chip hardware. This stripped-down variant of Codex is built for real-time coding, slashing roundtrip delays by 80% and speeding up first responses by 50%. ChatGPT Pro users get preview access today via the Codex app, CLI, and VS Code extension, while API access remains limited to select partners. The move signals OpenAI's pivot away from Nvidia-only infrastructure. If Spark delivers on its speed promise, expect coding assistants to feel less like autocomplete and more like a pair-programming partner who actually keeps up.
What happened
OpenAI just launched GPT-5.3-Codex-Spark, a blazingly fast coding model that churns out 1000+ tokens per second thanks to Cerebras' massive wafer-scale chip hardware. This stripped-down variant of Codex is built for real-time coding, slashing roundtrip delays by 80% and speeding up first responses by 50%. ChatGPT Pro users get preview access today via the Codex app, CLI, and VS Code extension, while API access remains limited to select partners. The move signals OpenAI's pivot away from Nvidia-only infrastructure.
Why it matters
If Spark delivers on its speed promise, expect coding assistants to feel less like autocomplete and more like a pair-programming partner who actually keeps up.