GPT-5.5 may burn fewer tokens, but it always burns more cash

OpenAI last month rolled out GPT-5.5, the latest iteration of its flagship frontier model family, alongside steep price hikes that have left developers and digital rights advocates questioning the true cost of advanced AI. Per 1 million tokens, GPT-5.5 now costs $5 for input, $0.50 for cached input, and $30 for output, doubling the rates of its predecessor GPT-5.4 across all tiers. The company claims these increases are offset by meaningful token efficiency gains, with the model delivering equivalent or better results using fewer tokens than earlier versions. "While GPT-5.5 is priced higher than GPT-5.4, it is both more intelligent and much more token efficient," the company said during the rollout.

Independent analysis from AI routing platform OpenRouter contradicts the extent of these savings. Their review of real-world usage data shows GPT-5.5 actual costs increased between 49 percent and 92 percent compared to GPT-5.4, even after factoring in efficiency improvements. Longer prompts, defined as those over 10,000 tokens, saw the most offset, with GPT-5.5 generating 19 percent to 34 percent fewer completion tokens. For shorter prompts under 10,000 tokens, completion lengths did not decrease meaningfully, leaving users with cost increases at the higher end of the range. "Our analysis shows that GPT-5.5 actual costs increased 49 percent to 92 percent," OpenRouter said. "Longer prompts, over 10k tokens, saw costs offset by shorter completions. Shorter prompts, under 10k, experience a higher cost increase where completions did not get shorter."

The pricing shift comes as OpenAI faces massive financial headwinds. Reports indicate the company expects to lose $14 billion in 2026, even as it plans to spend $50 billion on compute infrastructure this year alone. Rival Anthropic is in a similar position, projecting $11 billion in losses for 2026. Anthropic's recent Claude Opus 4.7 release did not include a public list price change, but OpenRouter found actual costs for prompts over 2,000 tokens increased 12 percent to 27 percent when cache absorption is accounted for, with only short prompts under 2,000 tokens seeing net savings from improved tokenization. "Our study of real Opus 4.7 usage shows that actual costs increased 12–27 percent for prompts above 2K tokens when cache absorption is taken into account," the biz said. "Short prompts under 2K were the exception, where significantly shorter completions offset the tokenizer overhead entirely."

Additional context from industry reports highlights related trends. Using AI via web interfaces, where users click through prompts and interactions, burns 45 times as many tokens as direct API usage, meaning casual users and non-technical adopters face far higher effective costs than enterprise API clients. OpenAI also faced criticism for restricting access to its GPT-5.5-Cyber variant to enterprise customers, a practice it previously slammed Anthropic for implementing. Meanwhile, Cloudflare announced plans to fire 1,100 staff whose roles were deemed replaceable by AI, underscoring the broader industry push to cut labor costs as compute spending soars.

Legal Basis: Data Protection Regulations in the AI Era

The financial pressures facing frontier AI firms carry direct implications for compliance with global data protection frameworks. OpenAI and Anthropic both process vast amounts of personal data from users across the globe, subjecting them to regulations including the European Union's General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA).

GDPR requires any organization processing personal data of EU residents to implement appropriate technical and organizational measures to protect that data, disclose processing practices transparently, and respond to user requests to access, delete, or port their data. Violations can result in fines of up to 4 percent of a company's global annual revenue or €20 million, whichever is higher. The GDPR Portal provides full details of these requirements. CCPA applies to companies serving California residents with more than $25 million in annual revenue, imposing similar transparency and opt-out obligations, with fines of up to $7,500 per intentional violation. More information is available via the California Department of Justice CCPA guide.

Upcoming legislation adds further compliance burdens. The EU AI Act classifies general-purpose AI models like GPT-5.5 as high-risk if they pose systemic risk, requiring providers to publish detailed technical documentation, implement robust data governance practices, and conduct third-party audits. Failure to comply with these rules could lead to additional fines of up to 7 percent of global revenue.

Impact on Users and Companies

Affected parties span individual users, third-party developers, AI firms, and regulators. For individual users, higher token costs translate to more expensive subscriptions for services like ChatGPT, or reduced access to advanced features as companies pass on costs. Developers integrating GPT-5.5 via API face 50 percent or higher cost increases for short prompts, forcing many to either raise prices for their own customers or cut spending in other areas.

Digital rights advocates warn that cost-cutting to offset AI losses often targets compliance teams and data protection infrastructure first. A startup using GPT-5.5 to power a health app, for example, might skip mandatory GDPR data audits to cover a 60 percent increase in AI API costs, putting sensitive user health data at risk of breach or misuse. OpenAI's decision to restrict GPT-5.5-Cyber to enterprise clients also limits access for smaller nonprofits and privacy-focused developers, who may turn to less regulated, lower-cost models with weaker data protection safeguards.

Regulators are also impacted, as cash-strapped AI firms become harder to monitor. Underfunded compliance teams are more likely to overlook data breach notifications, fail to respond to user data requests, or improperly retain user prompts and outputs, all of which trigger GDPR and CCPA penalties. For context, OpenAI's 2024 global revenue was estimated at $3.7 billion, meaning a maximum GDPR fine could reach $148 million per violation, far outstripping any short-term savings from cutting compliance spending.

What Changes Now

OpenAI has signaled further price increases for premium models are likely, as compute costs continue to rise faster than efficiency gains. The OpenAI API Pricing page confirms current rates, with no public commitment to cap future increases. OpenRouter's full analysis of GPT-5.5 and Opus 4.7 costs is available on their documentation site.

For users, the shift means auditing AI usage to prioritize API integrations over web interfaces, which burn far more tokens, and considering self-hosted open-source models to reduce reliance on proprietary systems that process personal data. Third-party companies should budget for rising AI costs now, and avoid diverting compliance funding to cover API bills, as regulatory fines would far exceed any short-term savings.

Regulators may also step up scrutiny of AI firms' financial health, requiring regular compliance audits for companies facing large losses to ensure data protection obligations are not sidelined. The EU AI Act's implementation in 2026, the same year OpenAI projects its $14 billion loss, will add additional oversight, with mandatory reporting for high-risk AI systems.

As frontier models grow more expensive to run, the tension between profit margins and user privacy will only intensify. For now, the cost of advanced AI is rising, and the bill may be paid not just in cash, but in eroded data protections for millions of users.