GPT-5.3-Codex-Spark: Real-Time Coding Model

TL;DR

• GPT-5.3-Codex-Spark is a smaller, faster model optimized for real-time coding — 1000+ tokens/sec
• Research preview available now for ChatGPT Pro in Codex app, CLI, and IDE extensions
• Not yet available in the API; use gpt-5.2-codex for API workflows

New

• GPT-5.3-Codex-Spark — Smaller, faster variant of GPT-5.3-Codex designed for near-instant responses in real-time coding scenarios. Delivers 1000+ tokens per second while maintaining capability for production code tasks.
• 128k context window — Text-only at launch with sufficient context for most coding workflows.
• Cerebras partnership — First milestone in collaboration to power faster inference.

How to Use

• CLI — Start a new thread with codex --model gpt-5.3-codex-spark or use /model during a session.
• IDE extension — Select GPT-5.3-Codex-Spark from the model selector in the composer.
• Codex app — Choose GPT-5.3-Codex-Spark from the model selector in the composer.

Important Notes

• Research preview limits — Usage has separate model-specific limits and doesn't count against standard Codex limits. During high demand, access may queue.
• API not yet supported — GPT-5.3-Codex-Spark is unavailable in the API at launch. Continue using gpt-5.2-codex for API workflows.
• Update required — If you don't see the model, update your CLI, IDE extension, or Codex app to the latest version.

To upgrade: npm install -g @openai/codex-cli@latest or check your IDE extension marketplace for updates.

Source: Codex