GPT-5.3-Codex-Spark

2026-02-12 · Publisher : OpenAI · Domain : Artificial Intelligence

OpenAI introduces GPT-5.3-Codex-Spark, a smaller version of GPT-5.3-Codex designed for real-time coding, served on Cerebras chips at more than 1,000 tokens per second.

What changes

First milestone of the partnership with Cerebras announced in January: the model runs on the Wafer Scale Engine 3.
Lightweight working style: targeted edits, no automatic test runs unless requested.
End-to-end latency improvements (streaming, inference stack, session initialization) that benefit all models.
Presented as the first of a family of ultra-fast models.

### Specifications - Context: 128,000 tokens, text only at launch. - Availability: research preview for ChatGPT Pro subscribers (Codex app, CLI, VS Code extension); API for a small number of partners. - Dedicated usage limits during the preview. - Identifier: gpt-5.3-codex-spark. ### Pricing Pricing not disclosed at launch (API access reserved for partners). ### Benchmarks Comparison with GPT-5.3-Codex on SWE-Bench Pro and Terminal-Bench 2.0 published as charts, with no numerical values in the announcement text. No usable numerical score. ### Key takeaways Spark targets interactive editing loops where latency matters as much as intelligence. It remains a preview reserved for Pro, with no public pricing.

GPT-5.3-Codex-Spark

What changes

Links