OpenAI relies on Cerebras to accelerate GPT-5.3-Codex-Spark

OpenAI has announced the release of GPT-5.3-Codex-Spark, the first model optimized for execution on Cerebras Systems' AI accelerators. This move represents a strategic diversification from the exclusive reliance on Nvidia and AMD for inference.

Cerebras Systems' accelerators stand out for their large chip surface area and high-speed on-chip memory, characteristics that make them particularly suitable for complex artificial intelligence workloads. GPT-5.3-Codex-Spark achieves a speed of 1,000 tokens per second (Tok/s) when running on these accelerators.

For those evaluating on-premise deployments, there are trade-offs to consider carefully. AI-RADAR offers analytical frameworks on /llm-onpremise to evaluate these aspects.