Introducing GPT-5.5 with NVIDIA's AI Researcher

The 10x speed figure comes from a single NVIDIA researcher in what is effectively a co-branded promotional video, not an independent benchmark or published evaluation. No methodology is given for how experiment execution time was measured, which makes the number impossible to verify or compare against anything.

This is the second GPT-5.5 launch video OpenAI published on the same day. The Perplexity-anchored release covered in 'Introducing GPT-5.5 with Perplexity' gave a concrete, testable claim: 56% token reduction on agentic workflows. This NVIDIA version offers a speedup figure without equivalent specificity, which suggests OpenAI is running a multi-partner announcement strategy where each video surfaces a different angle for a different audience. The autonomous code refactoring demo also connects directly to the Codex-based workspace agents OpenAI rolled out the same day, including the software review agent and IT ticket routing features. That timing is unlikely to be coincidental: the workspace agent rollout needs a capable underlying model to be credible, and GPT-5.5 is being positioned as that foundation.

If NVIDIA or OpenAI publish a reproducible benchmark showing the 10x figure across a defined task suite within the next 60 days, the claim earns serious weight. If no methodology surfaces, treat it as a marketing anchor number.