Codex Strategy Gains Speed in 2026
OpenAI has pushed its Codex plan further in 2026. Earlier this month, the company introduced GPT-5.3-Codex, giving priority to a coding-focused model instead of a general-purpose version for the first time.
Soon after the release, CEO Sam Altman described Codex’s growth as “insane.” According to him, usage increased by 50 percent in just one week. That rapid growth shows strong developer interest.
Now, the company has taken another step forward by launching its first real-time coding AI model, GPT-5.3-Codex-Spark.
GPT-5.3-Codex-Spark Enters Research Preview
OpenAI shared details about the new model in an official announcement. At present, ChatGPT Pro subscribers can access GPT-5.3-Codex-Spark in research preview.
The model works across:
- Codex app
- Command-line interface (CLI)
- IDE extension
Since this is an early version, some limits still apply. Over the next few weeks, the company plans to improve the system and expand availability.
Real-Time Code Generation at High Speed
The biggest highlight of Codex-Spark is its speed. The system can process up to 1,000 tokens per second, which allows it to write and edit code almost instantly.
Developers can create new code, fix errors, adjust logic, and improve interfaces in real time. A large 128,000-token context window helps the model manage regular and moderately complex tasks with ease.
For bigger and more complex projects, OpenAI still recommends using GPT-5.3-Codex.
Powered by Advanced Hardware
Speed improvements come from strong hardware support. OpenAI partnered with Cerebras to run Codex-Spark on the Wafer Scale Engine 3 AI accelerator.
This setup reduces delay and increases output speed. Because of that, developers receive responses much faster than before.
Performance Benchmarks Show Progress
Internal tests show that Codex-Spark performs better than GPT-5.1-Codex-mini on SWE-Bench Pro and Terminal-Bench 2.0. These tests measure software engineering ability.
Although GPT-5.3-Codex still delivers slightly stronger overall performance, Codex-Spark responds faster. That speed advantage makes it a major upgrade for real-time coding tasks.
API Access and Future Rollout
A small group of design partners will soon get API access. Through this step, OpenAI aims to learn how developers plan to use the model in real products.
Separate rate limits apply to Codex-Spark. Its usage does not affect standard limits. Wider access will roll out gradually in the coming weeks.