OpenAI hikes GPT‑5.5 per‑token prices up to 2×

OpenAI raised the sticker price on its newest flagship API model, and the increase is not subtle. GPT‑5.5 now costs 2× GPT‑5.4 on both input and output tokens, which means teams that upgrade blindly can watch inference bills jump fast. The awkward part is that GPT‑5.5 is also supposed to be more efficient on some tasks. So the real question is not “is it better?” It’s “does better save enough tokens to pay for itself?” ### What exactly changed? The cleanest comparison is on OpenAI’s own pricing table. GPT‑5.5 is listed at $5 per 1M input tokens, $0.50 cached input, and $30 per 1M output tokens. GPT‑5.4 sits at $2.50 input and $15 output. So for standard text usage, the new model is simply twice as expensive per token before any efficiency gains enter the picture. ### Is every GPT‑5.5 call priced the same? (openai.com) No — and this is where long-context users need to pay attention. OpenAI says prompts above 272,000 input tokens trigger higher session pricing for GPT‑5.5: 2× input and 1.5× output for the full session. That means the model is not just pricey in the ordinary sense; it gets even pricier when you use the giant context window that makes frontier models attractive for research, codebase analysis, and agent workflows. ### Doesn’t GPT‑5.5 use fewer tokens? Sometimes, yes. That is the whole economic argument for the upgrade. OpenAI pitches GPT‑5.5 as a stronger model for coding and professional work, and outside measurements cited by The Register say it can generate 19% to 34% fewer completion tokens on longer prompts. But a 19% to 34% token reduction does not cancel out a 100% per-token price increase. Basically, lower usage can soften the hit, but it usually does not erase it. (developers.openai.com) ### Where does the math still work? It works when model quality changes downstream costs. If GPT‑5.5 writes cleaner code, needs fewer retries, avoids tool-call loops, or reduces human review time, then the expensive tokens may still be cheaper at the job level. That matters most in agentic coding and professional workflows, where one bad answer can trigger more model calls, more tool execution, and more employee time. (openai.com) OpenAI is clearly aiming GPT‑5.5 at exactly those use cases. ### Who gets hit hardest? Heavy API users with long prompts and lots of output. Think coding copilots, document-heavy enterprise assistants, legal or research systems, and anything that shoves giant context windows into the model. The cached-input discount helps repeat context, but it does not solve the output side, and output is where GPT‑5.5’s list price really bites at $30 per 1M tokens. (openai.com) ### What about the pro model? GPT‑5.5 Pro is a different story — and a warning about where premium pricing may be heading. OpenAI lists GPT‑5.5 Pro at $30 per 1M input tokens and $180 per 1M output, with no cached-input discount. That is not a casual default model. It is a specialist tool for cases where accuracy or capability matters more than cost discipline. (openai.com) ### So what should buyers do now? Benchmark whole tasks, not token rates in isolation. Measure retries, latency, tool calls, review time, and output length. A better model can be cheaper in practice — but only if it changes the workflow enough to offset the higher meter. With GPT‑5.5, OpenAI is making that tradeoff explicit. The best model is no longer automatically the economical one. (developers.openai.com) ### Bottom line GPT‑5.5 looks like a classic frontier-model deal: more capability, more cost, and more need for disciplined testing. If you run short, high-value tasks, the upgrade may be easy to justify. If you run long-context systems at scale, the price jump is real — and now large enough that finance teams will notice. (openai.com)