GPT-5.5 tops benchmarks but still hallucinates frequently and costs 20 percent more over the API

The more pointed issue isn't the benchmark result — it's that OpenAI is raising API prices at the same moment it's shedding product teams and consolidating around enterprise. A 20 percent cost increase on top of unresolved hallucination rates is a real ask for developers who are already scrutinizing per-token spend.

This fits directly into the pattern visible in our April 17 coverage of OpenAI's organizational contraction: the departures of Kevin Weil and Bill Peebles, the discontinuation of Sora, and the consolidation toward Codex and enterprise tooling all point to a company tightening its cost structure while pushing pricing upward on the API side. The tokenmaxxing piece from the same week is also relevant here — developers already warned that optimizing token usage creates hidden costs and maintenance burden, and a 20 percent price hike compounds that pressure. What OpenAI is effectively doing is betting that benchmark leadership justifies the premium even as the hallucination problem remains unsolved.

Watch whether Anthropic or Google respond with a competing price cut or a targeted reliability benchmark in the next 60 days. If either does, it signals that the 20 percent increase is a vulnerability OpenAI's competitors are willing to exploit directly.