We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
OpenAI Blog·Feb 12·5 min read
Introducing GPT-5.3-Codex-Spark
TL;DR
Prior coding assistants generated slowly and handled limited context, creating friction in real-time development workflows. OpenAI built GPT-5.3-Codex-Spark, a real-time code generation model with 15x faster inference and 128k token context window (the amount of text it can process at once).
✦ Why It Matters
Engineers can now use real-time AI code generation without disruptive delays, accelerating development velocity and reducing context-switching friction.
Key Takeaways
- Evaluate GPT-5.3-Codex-Spark for latency-sensitive coding workflows requiring immediate feedback.
- Leverage 128k context window to reference entire files or multi-file projects in single prompts.
- Test real-time code generation in ChatGPT Pro to identify integration opportunities in your IDE.
- Monitor inference speed improvements to quantify productivity gains in your development process.
Full Summary
Coding assistants have traditionally suffered from latency and context limitations that disrupt developer workflows. GPT-5.3-Codex-Spark addresses these constraints through architectural optimizations enabling 15x faster token generation compared to prior versions, while expanding the context window to 128k tokens—allowing the model to reference and understand larger codebases and longer conversation histories simultaneously.
The model employs real-time inference techniques to minimize latency during interactive coding sessions. Performance gains were measured against baseline generation speed and context retention benchmarks.
This advancement enables developers to use AI assistance more fluidly during active coding, reducing wait times and supporting more complex, multi-file programming tasks. The research preview rollout to ChatGPT Pro users provides early feedback before broader deployment.