OpenAI clocks GPT-5.3-Codex-Spark at 1,200+ tokens per second
OpenAI just boosted the inference speed of GPT-5.3-Codex-Spark by roughly 30%, pushing throughput past 1,200 tokens per second. That blistering output makes real-time code generation and autonomous developer workflows feel virtually instantaneous. OpenAI says broader speed upgrades across its model lineup are coming next.
Waiting for a coding model to stream tokens one line at a time is the ultimate flow-state killer.
OpenAI just gave its developer tools a massive adrenaline shot. OpenAI team member Thomas Sottiaux announced on X that GPT-5.3-Codex-Spark is now roughly 30% faster, pushing output speeds past a blistering 1,200 tokens per second.
Blink and the code is done
At 1,200 tokens per second, entire functions and boilerplate scripts appear almost instantly the moment you hit enter.
While OpenAI didn't share the exact technical magic behind the optimization, they did tease that broader speed improvements across their lineup are on the way.
Why it matters
For developers, latency is everything. If an AI assistant takes two seconds to generate an answer, you lose your train of thought.
Crossing the 1,200 tokens-per-second threshold effectively turns generative AI from an awkward assistant into real-time muscle memory inside your IDE. It puts serious pressure on rivals to optimize inference efficiency rather than just chasing benchmark scores. Speed is becoming the main battleground for everyday developer tools.
Sources (1)
- @thsottiaux(opens in a new tab)Originalx.com