OpenAI Launches GPT-5.3-Codex-Spark

Why it matters

- The Cerebras WSE-3 chip is one of the largest processors ever built, featuring 4 trillion transistors and 900,000 AI-optimized cores on a single dinner-plate-sized wafer of silicon. This design provides 125 petaflops of peak AI performance. - This model marks a significant hardware shift for OpenAI, representing the first major model from the company to be built on Cerebras chips rather than Nvidia GPUs. - The original OpenAI Codex, based on GPT-3, was first introduced in August 2021 and provided the underlying technology for the initial version of GitHub Copilot. - OpenAI had previously deprecated its legacy Codex models in March 2023, transitioning developers to newer GPT-3.5 and GPT-4 class models before reintroducing the "Codex" name for a more advanced coding agent in 2025. - Recent high-performing coding models from OpenAI have set a high bar, with the `o3` reasoning model solving nearly 72% of issues in a GitHub-based coding benchmark, a significant increase from the 4.4% achieved by the best models in late 2023. - The ChatGPT Pro subscription required to access the model costs $200 per month and is targeted at researchers, developers, and other power users requiring unlimited access to OpenAI's top-tier models.

Key numbers

• OpenAI launched GPT-5.3-Codex-Spark, a new AI model designed for real-time coding.
• Built using Cerebras’ WSE-3 chips, the model reportedly completes tasks 15 times faster than its predecessors.
• - The Cerebras WSE-3 chip is one of the largest processors ever built, featuring 4 trillion transistors and 900,000 AI-optimized cores on a single dinner-plate-sized wafer of silicon.
• This design provides 125 petaflops of peak AI performance.

Sources

• OpenAI launched
• The Cerebras WSE-3
• This design provides
• This model marks a
• The original OpenAI
• Recent high-performing
• The ChatGPT Pro subscription

Quick answers