Double the price on paper and in practice
OpenAI introduced the GPT-5.5 model on April 23, 2026, with an ambitious goal: to push the boundaries of agentic artificial intelligence for real work. Alongside significant benchmark leaps — such as 82.7% on Terminal-Bench 2.0 or 78.7% on OSWorld-Verified — came news that made more than one CTO freeze. The price per million input tokens rose from $2.50 to $5.00, and the price for output tokens from $15 to $30. That means double compared to GPT-5.4.
For the most demanding tasks, OpenAI also offers the GPT-5.5 Pro variant, with an API price of $30 per million input tokens and $180 per million output tokens. That's six times the input price of standard GPT-5.5 and twelve times that of GPT-5.4. For this money, users get even higher accuracy, but for many companies this may be financially unsustainable.
OpenAI argues that GPT-5.5 is "significantly more token-efficient" and that for most users in ChatGPT and Codex, it means better results with a lower overall token count. However, this claim depends on the type of task — and that is exactly what the OpenRouter aggregator independently verified.
What the OpenRouter analysis showed
OpenRouter conducted a controlled study on a group of users who switched from GPT-5.4 to GPT-5.5. The comparison took place on the same workflows, with the same users, and in the same period (April 2026). The key finding is that real costs rose by 49% to 92%, even though the model sometimes generates shorter responses.
The table below shows the average price per million OpenRouter tokens by prompt length:
| Prompt length | GPT-5.4 ($/M) | GPT-5.5 ($/M) | Increase |
|---|---|---|---|
| < 2K tokens | $4.89 | $9.37 | +92% |
| 2K – 10K | $2.25 | $3.81 | +69% |
| 10K – 25K | $1.42 | $2.15 | +51% |
| 25K – 50K | $1.02 | $1.65 | +62% |
| 50K – 128K | $0.74 | $1.10 | +49% |
| 128K+ | $0.71 | $1.31 | +85% |
Short queries were hit the hardest. For prompts shorter than 2,000 tokens, GPT-5.5 even generates 7% longer responses than GPT-5.4, which, combined with the doubled price, leads to an almost doubled bill. On the other hand, for long prompts over 10,000 tokens, the model actually saves — generating 19% to 34% fewer tokens, which partially compensates for the higher per-token price.
How GPT-5.5 stands in the competition
OpenAI's pricing policy cannot be judged in isolation. Anthropic offers the Claude Opus 4.7 model for $5 per million input tokens and $25 per million output tokens. This means GPT-5.5 is equally expensive on input but 20% more expensive on output. For comparison: Google's Gemini 3.1 Pro is traditionally cheaper in a similar performance class, although specific prices vary by region and deployment type.
Performance-wise, GPT-5.5 holds strong: in coding, it outperforms Claude Opus 4.7 on Terminal-Bench 2.0 (82.7% vs. 69.4%), and in agentic tasks it leads on GDPval (84.9% vs. 80.3%). The question for companies therefore is: is the performance difference worth the price? For teams using short interactive queries, maybe not. For developers of agentic systems with long context, probably yes.
What this means for Czech companies and developers
For the Czech market, there are two key pieces of information. First: GPT-5.5 is available as part of the ChatGPT Plus subscription (approx. 600 CZK per month) and ChatGPT Pro (from approx. 3,000 CZK per month) — so for regular users, the subscription price remains unchanged. However, for companies using the API, costs rise sharply.
Second: OpenAI guarantees GDPR compliance for enterprise customers and offers the option of EU data residency. This is crucial for Czech companies processing personal data. Business and Enterprise plans also bring advanced administrative tools and SLA.
In practice, this means that small and medium-sized Czech companies should carefully consider whether they really need the full power of GPT-5.5, or whether the cheaper GPT-5.5 Thinking Mini or competing models will suffice. For startups and teams with limited budgets, it may be reasonable to stay with GPT-5.4 for routine tasks and deploy GPT-5.5 only where its agentic capabilities deliver measurable value.
Deployment recommendations
Based on data from OpenRouter, a simple rule can be formulated: the shorter the queries you send, the more GPT-5.5 will hurt you. If your product relies on thousands of short API calls, switching to GPT-5.5 will double or nearly double costs. On the other hand, if you work with long documents, code, or complex agentic workflows, the higher per-token price is partially offset by shorter responses.
OpenAI offers Batch and Flex processing at half price, which may be an interesting path for non-interactive tasks. For critical systems, Priority processing at 2.5x the price is available — but that's already a segment where we're talking about costs in the hundreds of dollars per million tokens.
Should I switch from GPT-5.4 to GPT-5.5 if I use API for a chatbot?
If your chatbot handles short queries (under 2,000 tokens), the switch will nearly double costs without significantly shortening responses. Consider A/B testing with a smaller portion of traffic and compare cost per conversion, not just cost per token.
What is the difference between GPT-5.5 and GPT-5.5 Pro in practice?
GPT-5.5 Pro costs six times more on input and six times more on output compared to standard GPT-5.5. OpenAI recommends it for the most complex tasks requiring maximum accuracy. For ordinary business purposes, the standard version is sufficient; Pro is more for scientific research, financial modeling, or legal analysis.
Is GPT-5.5 also available for Czech users outside of API?
Yes. GPT-5.5 Thinking is part of the ChatGPT Plus subscription, which is commonly available in the Czech Republic for approx. 600 CZK per month. The GPT-5.5 Pro variant is available to Pro, Business, and Enterprise subscribers. For developers, the API is available globally, including the Czech Republic.
