GPT‑5.3‑Codex‑Spark is OpenAI’s research-preview model built for real-time coding in Codex. Designed for near-instant iteration, it can generate code at very high speed (reported as over 1,000 tokens per second) and supports longer-context workflows, helping developers iterate faster in the Codex app, CLI and IDE tools.

“AI coding assistants” are everywhere — but most still feel like a chat interface bolted onto development work. OpenAI’s GPT‑5.3‑Codex‑Spark is different: it’s designed for real-time coding iteration, where responsiveness matters as much as capability.

In February 2026, OpenAI released Codex‑Spark as a research preview model aimed at making AI-assisted coding feel near-instant, with OpenAI describing performance of more than 1,000 tokens per second when served on low-latency hardware.

Key benefits at a glance

Near-instant generation for tighter feedback loops

Codex‑Spark is optimised to keep pace with live coding: quick edits, rapid retries, and fast back-and-forth when you’re debugging or refactoring. OpenAI positions it as its first model designed specifically for real-time coding.

Longer-context coding workflows

If you’re working across large files, multiple modules, or complex diffs, longer context matters. OpenAI’s research listing for Codex‑Spark highlights 128k context, which supports more “whole-project” style iteration than typical chat-based coding help.

Research preview access for Pro users

Codex‑Spark is currently offered as a research preview and is available to ChatGPT Pro users through Codex surfaces, including the app and CLI.

Updated as of 13 February 2026: Codex‑Spark is listed by OpenAI as a research preview model for real-time coding.

How it works (what’s actually new)

Codex‑Spark is described as a smaller version of GPT‑5.3‑Codex, tuned for speed so you can iterate quickly during active development — rather than waiting for long generations to complete. OpenAI also frames it as the first milestone in its partnership with Cerebras, which focuses on ultra-low-latency serving for fast inference.

The practical impact is simple: the model is built to support new interaction patterns — more like a responsive pair programmer than a slow “generate-and-paste” tool.

Practical steps: how to try GPT‑5.3‑Codex‑Spark

1. Confirm access (ChatGPT Pro): Codex‑Spark is listed under Pro benefits as research preview access.

2. Use it where you code: OpenAI lists Codex‑Spark availability across Codex tooling, including the Codex app, CLI, and IDE extension.

3. Switch the model in your workflow: OpenAI’s Codex docs show model selection for the CLI (e.g., using gpt-5.3-codex-spark) and describe using the faster model for responsive tasks.

Where Codex‑Spark fits best

Codex‑Spark is most compelling when latency is the bottleneck:

• tight edit–run–fix loops

• fast refactors across a codebase

• rapid testing of small implementation changes

• “pair programming” style collaboration in an IDE

For deeper multi-step engineering work, teams may still prefer heavier models — but Spark is built for the moments where speed changes how you work.

Summary

GPT‑5.3‑Codex‑Spark is OpenAI’s first model designed for real-time coding, offered as a research preview for ChatGPT Pro users. If your goal is faster iteration — not just occasional code generation — Spark is worth testing in your daily workflow.

Next steps: If you’re exploring how tools like Codex fit into your engineering workflow (governance, enablement, best-practice prompts, and safe rollout), contact Generation Digital for guidance.

FAQs

What is GPT‑5.3‑Codex‑Spark?
GPT‑5.3‑Codex‑Spark is OpenAI’s research-preview model designed for real-time coding in Codex, optimised for near-instant iteration.

How is Codex‑Spark different from GPT‑5.3‑Codex?
OpenAI describes Codex‑Spark as a smaller version of GPT‑5.3‑Codex tuned for speed and responsiveness, intended to support new real-time coding interaction patterns.

How fast is Codex‑Spark?
OpenAI states it can deliver more than 1,000 tokens per second when served on ultra-low-latency hardware. Some OpenAI materials and coverage also describe “15x faster generation,” but the most consistent official metric is the 1,000+ tokens/sec framing.

Who can access GPT‑5.3‑Codex‑Spark?
OpenAI lists Codex‑Spark as a research preview available to ChatGPT Pro users (via Codex tooling such as the app and CLI).

Where can I use it?
OpenAI documentation references Codex‑Spark usage in the Codex app, Codex CLI, and an IDE extension.