Introducing GPT-5.5 with Perplexity

The 56% token reduction figure is presented without a stated baseline: 56% fewer tokens than GPT-5, than GPT-4o, or than some internal benchmark run? That denominator matters enormously for evaluating whether this is a genuine architectural improvement or a prompt-efficiency tuning pass dressed up as a model release.

Modelwire has no prior coverage to anchor this to directly, so this sits in a broader pattern worth naming: OpenAI has accelerated its cadence of named model variants (GPT-4o, GPT-4.5, now GPT-5.5) in ways that make version-to-version comparisons increasingly difficult for developers to track. The Perplexity co-sign is doing real work here as a credibility signal, but Perplexity has a commercial relationship with OpenAI's API, which makes them a motivated rather than independent validator. The agentic workflow framing also echoes a consistent OpenAI messaging thread around Codex and operator-facing products, though without prior coverage we cannot say whether this announcement materially advances that roadmap or restates it.

Watch whether third-party developers running high-volume agentic pipelines (outside the Perplexity relationship) publish reproducible token-count comparisons against GPT-5 on identical task sets within the next 60 days. If those numbers don't land near 50%, the 56% figure is likely task-specific, not general.