In an exciting leap forward in the landscape of artificial intelligence, OpenAI has unveiled its latest model, GPT-5.3-Codex-Spark. This cutting-edge AI framework marks a significant shift in the company’s hardware strategy by utilizing Cerebras wafer-scale chips as opposed to the traditional Nvidia GPUs. Designed for enhanced throughput and low-latency performance, Codex-Spark aims to deliver a real-time, interactive coding experience, catering primarily to developers who need rapid coding assistance.
We’re sharing Codex-Spark on Cerebras as a research preview to ChatGPT Pro users so that developers can start experimenting early while we work with Cerebras to ramp up datacenter capacity, harden the end-to-end user experience, and deploy our larger frontier models.
One of the standout features of Codex-Spark is its impressive throughput, enabling it to run at approximately 1,000 tokens per second. This marks an astonishing 15 times improvement over previous versions, significantly augmenting the live coding assistance experience. OpenAI claims that the model has been meticulously crafted for real-time interaction, allowing developers to make targeted edits, refine logic, and reshape interfaces while observing immediate results.
The emphasis on low latency means that Codex-Spark is optimized for interactive coding workflows rather than deep reasoning tasks. It remains capable of handling long-running processes, continuing to operate effectively for hours, days, or even weeks without the need for intervention. This blend of speed and reliability makes it an ideal tool for software developers working on complex coding projects.
Performance evaluations conducted on benchmark tests like SWE-Bench Pro and Terminal-Bench 2.0 have shown that Codex-Spark delivers results between GPT-5.1-Codex-mini and GPT-5.3-Codex, all while achieving these milestones in a fraction of the time previously required. Moreover, OpenAI has prioritized end-to-end improvements across their entire request-response pipeline, ensuring lower latency benefits for all models in the family.
Under the hood, we streamlined how responses stream from client to server and back, rewrote key pieces of our inference stack, and reworked how sessions are initialized so that the first visible token appears sooner and Codex stays responsive as you iterate.
Further enhancements include the introduction of a persistent WebSocket connection and multiple optimizations within the Responses API. These updates are pivotal as they reduce per client/server roundtrip overhead by a staggering 80%, cut down per-token processing time by 30%, and halve the time taken to spit out the first token. OpenAI’s commitment to maintaining these rapid response times will consequently benefit all their models going forward.
Codex-Spark operates on Cerebras’ Wafer Scale Engine 3 accelerators, which are excellently tailored for low-latency and high-speed inference applications. However, OpenAI reassures users that this does not signify a complete transition away from GPUs within their training and inference pipeline. Instead, Cerebras accelerators are designed to complement GPUs to achieve optimal performance across different scenarios.
The announcement of Codex-Spark has incited lively discussions across online forums. Some Reddit users expressed a preference for prioritizing “maximum intelligence and reliability” rather than sheer speed. User Tystros pointed out that if improved results are achievable with a longer processing time, they would take that route, stating, “[if the results are better when it takes one hour to complete a task, I happily wait one hour].” Another user, stobak, highlighted the cumulative costs associated with rapid iterations of quicker models, sparking insightful debates about the trade-offs of speed versus quality.
While many have embraced the new speed improvements, some commentators have expressed caution regarding the drastic speed claims. Nicholas Van Landschoot noted on X.com that, in practical benchmarks, the speed improvements appear closer to 1.37 times rather than the advertised 15 times. He observed that the impressive speed claim stems from benchmarks against a specific configuration of Codex, referred to as x-high, which emphasizes longer reasoning times for improved accuracy.
Codex-Spark is equipped with a 128k context window and supports text-only input for the moment. OpenAI is already contemplating the introduction of faster models featuring larger contexts based on insights gathered from the developer community’s usage patterns. Such developments indicate a persistent commitment to innovation and responsiveness to user needs within the rapidly evolving AI landscape.
Inspired by: Source
