OpenAI presents GPT-5.3-Codex-Spark, its first model capable of encoding in real time with “ultra-fast” responses

OpenAI has presented GPT-5.3-Codex-Sparkits first model designed for real time coding with capabilities to execute its results “almost immediately”, delivering more than 1,000 tokens per second.

A week after unveiling GPT-5.3-Codex, its most advanced coding artificial intelligence (AI) that helped in its own creation, the technology company led by Sam Altman has unveiled a new reduced version designed for real-time coding.

It is about GPT-5.3-Codex-Spark and, as OpenAI explained in a statement on its blog, it allows you to make specific edits, redefine the logic or perfect the interfaces when carrying out coding tasks, offering “immediate results” and real-time interaction.

That is, users can collaborate with the model interrupting you or modifying requests while you workalso receiving almost immediate responses. However, it must be taken into account that, to achieve this speed, the model makes minimal and specific edits, and does not run tests automatically unless specified.

As a result, users will be able to use Codex-Spark to carry out both more labor-intensive, long-term tasks and get work done on the spot. For this, the model has been presented with a 128K context window, text only.

The technology company has also highlighted that Codex-Spark demonstrates “excellent performance” in the SWE-Bench Pro and Terminal-Bench 2.0 benchmarks, which evaluate software engineering capabilities. In fact, according to these tests, it is capable of complete tasks “in much less time” than GPT-5.3-Codex.

It must be taken into account that, to achieve this speed in Codex-Spark while ensuring efficient results, OpenAI has detailed that this model is part of its collaboration with the company Brainsso it runs on the motor Wafer Scale Engine 3 from the company, an AI accelerator designed specifically for high-speed inference, which helps prioritize latency.

With all this, Codex-Spark has been launched as a research preview for users of ChatGPT Pro in the latest versions of the Codex appso they can start experimenting with this technology.

OpenAI has also warned that since the model runs on specialized low-latency hardware, it will have an independent speed limit that can be adjusted based on demand during the research trial.