GPT-5.5 costs 49 to 92 percent more than its predecessor, depending on the input length

The 49-to-92 percent range is the tell: that spread means the pricing model is specifically punitive for long-context enterprise workloads, which are exactly the use cases most likely to drive high-volume contracts. Buyers optimizing for throughput at scale are absorbing the largest increases, not casual API users.

Anthropic's simultaneous price increase on Opus 4.7 is the more important data point here, and it connects directly to the sycophancy research covered from Simon Willison's site in early May. That piece exposed how Claude's behavioral alignment is uneven across domains, yet Anthropic is raising prices anyway, suggesting the labs believe enterprise demand is inelastic enough to absorb cost increases regardless of unresolved capability questions. The broader pattern is that both OpenAI and Anthropic are tightening margins on their most capable models at the same time, which looks less like independent pricing decisions and more like a coordinated read on what the market will bear before IPO windows open.

Watch whether any major cloud reseller (AWS Bedrock, Azure AI Foundry, Google Cloud) absorbs part of the markup through negotiated volume pricing in the next 60 days. If they do, it signals the list price increases are partly theatrical. If they pass them through unchanged, enterprise migration pressure toward open-weight alternatives becomes a real near-term story.