AI Research
Feb 12, 2026
OpenAI launches GPT-5.3-Codex-Spark model on Cerebras chips, achieving high coding speeds
Feb 12, 2026
AI Summary
OpenAI has introduced its GPT-5.3-Codex-Spark coding model, which operates on Cerebras hardware instead of Nvidia. This model is capable of processing over 1,000 tokens per second, significantly outpacing its predecessor and offering new capabilities for developers.
- OpenAI released the GPT-5.3-Codex-Spark coding model on Cerebras chips, marking its first production model not reliant on Nvidia hardware.
- The new model processes code at over 1,000 tokens per second, approximately 15 times faster than its predecessor.
- In comparison, Anthropic's Claude Opus 4.6 reaches about 2.5 times its standard speed of 68.2 tokens per second but is a larger model.
- Sachin Katti, head of compute at OpenAI, expressed enthusiasm for the partnership with Cerebras and the introduction of fast inference capabilities.
- Codex-Spark is currently available as a research preview for ChatGPT Pro subscribers and will have API access for select design partners.
- The model features a 128,000-token context window and is text-only at launch.