OpenAI launches GPT-5.3-Codex-Spark, the first real-time coding model
Next step
Open the bot or continue inside this section.
Article -> plan in AI
Paste this article URL into any AI and get an implementation plan for your project.
Read this article: https://vibecode.morecil.ru/en/%D0%BD%D0%BE%D0%B2%D0%BE%D1%81%D1%82%D0%B8/openai-%D0%B2%D1%8B%D0%BF%D1%83%D1%81%D1%82%D0%B8%D0%BB%D0%B0-gpt-5-3-codex-spark--%D0%BF%D0%B5%D1%80%D0%B2%D1%83%D1%8E-%D0%BC%D0%BE%D0%B4%D0%B5%D0%BB%D1%8C-%D0%B4%D0%BB%D1%8F-%D1%80%D0%B5%D0%B0%D0%BB%D1%8C%D0%BD%D0%BE%D0%B3%D0%BE-%D0%B2%D1%80%D0%B5%D0%BC%D0%B5%D0%BD%D0%B8-%D0%B2-%D0%BA%D0%BE%D0%B4%D0%B8%D0%BD%D0%B3%D0%B5/
Work in my current project context.
Create an implementation plan for this stack:
1) what to change
2) which files to edit
3) risks and typical mistakes
4) how to verify everything works
If there are options, provide "quick" and "production-ready". How to use
- Copy this prompt and send it to your AI chat.
- Attach your project or open the repository folder in the AI tool.
- Ask for file-level changes, risks, and a quick verification checklist.
February 12, 2026 – OpenAI announced the launch of the research preview model GPT-5.3-Codex-Spark. This is a smaller version of GPT-5.3-Codex, specially designed for instant interaction with the developer.
The main thing is 30 seconds
- More than 1,000 tokens per second on Cerebras hardware
- 128k context as long as the text
- Available for all ChatGPT Pro users
- The first model in the line of "ultra-fast" Codex models
Why does a developer need it? n
Until now, Codex has been great at long tasks (hours and days of battery life), but in real time felt "reflective."
Spark changes this completely: you can now interrupt the model, instantly change the logic, rebuild the interface and immediately see the result.
Codex-Spark is the first step towards two modes of operation: long horizon + instant iteration. Over time, they will merge into one seamless experience, the official announcement reads.
Results on benchmarks
SWE-Bench Pro*
Spark shows accuracy close to the large model, but performs tasks many times faster.
Terminal-Bench 2.0
- GPT-5.3-Codex-Spark → 58.4%
- GPT-5.3-Codex > 77.3% (but much slower)
- GPT-5.1-Codex-mini 46.1%
Technical improvements to latency
OpenAI didn’t just make a fast model — they completely rebuilt the pipeline
- 80% decrease in overhead for every roundtrip
- 30% less overhead on token
- 50% faster than time-to-first-token
All this is possible thanks to a constant WebSocket connection and optimizations of the Responses API. All other Codex models will receive these improvements soon.
Partnership with Cerebras
It is powered by the Wafer Scale Engine 3 (Cerebras). This is the first time OpenAI has used a dedicated low-latency accelerator in parallel with its GPU clusters.
“What excites us most is the new patterns of interaction that become possible at this speed,” said Sean Lie, CTO and co-founder of Cerebras.
How can I try it
Today, Spark is available at:
- Codex app (latest version)
- Codex CLI
- VS Code Extension
The model has separate rate limits (so as not to burden the main infrastructure). With high demand, queues are possible.
What next
OpenAI is the first model in the ultra-fast family. Plans:
- large-scale
- multimodality
- longer context
- smooth switching between “long” and “instant” modes
Conclusion for Vibcoders
Whereas Codex used to be a “smart assistant who thinks sometimes,” it’s now a real partner who reacts faster than you can finish the next line.
For those who already work with Codex every day, this is one of the most notable jumps in the last year and a half