x.com/scaling01/status/2096228452512747587
1 correction found
for OpenAI we pretty much know that what we see publicly is only ~1-2 months behind what they have internally
OpenAI’s own docs provide a clear counterexample: GPT-5.3-Codex was already deployed internally five months before its public launch, so the public/internal gap is not generally just 1–2 months.
Full reasoning
This claim overstates how small OpenAI’s internal-to-public lag is.
A direct counterexample appears in OpenAI’s own safety materials for GPT-5.3-Codex:
- OpenAI’s Deployment Safety Hub says: “We began asynchronously monitoring internally deployed models such as GPT-5.3-Codex five months ago.”
- OpenAI’s launch materials say GPT-5.3-Codex was publicly launched on February 5, 2026.
Taken together, that means GPT-5.3-Codex had already been internally deployed about five months before its public release — much longer than the post’s claimed “~1-2 months.”
Because the post states this as something we “pretty much know” about OpenAI generally, a single official counterexample is enough to show the claim is not reliable as stated.
2 sources
- GPT-5.3-Codex System Card - OpenAI Deployment Safety Hub
We began asynchronously monitoring internally deployed models such as GPT-5.3-Codex five months ago.
- Introducing GPT-5.3-Codex | OpenAI
February 5, 2026 ... Today, we launched GPT-5.3-Codex, our most capable agentic coding model yet.