unknown

VibeCoders of the world, I only have one thing to say: OpenAI just built a coding monster.

The speed jump in gpt-5.3-codex xhigh is massive. It was likely boosted by the OpenAI x Cerebras partnership, plus whatever the OpenAI Codex team cooked behind the scenes. 

And now that we can run multiple multi-agent setups with real depth subagents calling subagents Codex has turned into an absolute beast for shipping high-quality software.

Yes, you will burn through your limits much faster. The 2x until April makes it easier for now, but after that I honestly worry the cap will evaporate quickly. Still, Codex’s biggest pain point slow implementation of larger, more complex features finally feels solved.

Congrats to @OpenAI @OpenAIDevs  @sama @gdb and the Codex team @thsottiaux  @embirico. This is insane.

PS I’m running seven Pro accounts to squeeze every drop out of the doubled limit until April

Tibo

A brief clarification that only the Spark model is served through Cerebras infrastructure, while the speed optimizations for GPT-5.3-Codex come from a different source, with hints that further speed improvements are still on the way.

Only the spark model is served through Cerebras, all speed optimizations for GPT-5.3-Codex are something different and the best is yet to come speed-wise! https://t.co/BsYqWpirU5

<p>Only the spark model is served through Cerebras, all speed optimizations for GPT-5.3-Codex are something different and the best is yet to come speed-wise! https://t.co/BsYqWpirU5</p>