OpenAI’s new API is 200x more expensive than competition

OpenAI’s newly launched 01 Pro API pricing lands at $150 per million input tokens and 600 output tokens—an order-of-magnitude jump that makes it dramatically more expensive than competing reasoning models. The core issue isn’t just sticker shock; it’s that the “reasoning” work behind the scenes consumes extra tokens, so users pay for compute that often doesn’t translate into better results or better ergonomics.

The transcript breaks down how token-based billing turns “thinking” into cost. Tokens are the small chunks models process to predict the next token in a sequence, and more tokens generally means more GPU work. With reasoning-enabled models, the model can generate additional intermediate tokens before producing the final answer. OpenAI doesn’t expose those thinking tokens in the API, but they still exist—so enabling higher-effort modes can inflate both input and output token counts even when the final response length looks similar.

That dynamic helps explain why cheaper models can feel faster and more cost-effective. The speaker contrasts OpenAI’s pricing with Google’s Gemini 2.0 Flash, which undercuts older expectations by combining competitive quality with speed and low cost—attributed to Google’s vertical integration (its own processors, architecture, training data pipeline, and infrastructure). The transcript also points to competitive pressure from models like DeepSeek’s reasoning releases, which have repeatedly forced price adjustments across the market.

Within OpenAI’s own lineup, the pricing story becomes more contentious. The speaker says 03 Mini High delivers better real-world outcomes than 01 Pro while costing far less, and that 01 Pro’s experience is not only expensive but also unreliable in practice—failing requests, awkward UI behavior, and even incorrect answers on a known Advent of Code-style test. Even when 01 Pro eventually returns something, the transcript frames it as a poor trade: higher cost, slower or brittle behavior, and results that don’t justify the premium.

A key quantitative claim anchors the argument: 01 Pro is about 136x more expensive than 03 mini high for input tokens and roughly the same multiple for output tokens (600 vs 440). The speaker interprets this as a pricing outlier—possibly a “feature parody” move to ship something tied to internal work rather than a model optimized for value. They also argue that OpenAI’s margins on heavy usage models can be thin or negative, citing the earlier example of OpenAI’s 01 Pro being offered on a $200/month ChatGPT tier that allegedly cost money due to heavy usage.

The transcript ends by reframing the market direction: while OpenAI’s API pricing jumps upward, other providers keep pushing down costs and improving efficiency. The speaker expresses confusion that OpenAI would release a model that is both more expensive and, in their testing, worse than 03 Mini High—then suggests that the “high” tiers may be better understood as an allowed “effort” budget rather than a fundamentally different model, with UI limitations forcing separate naming on OpenAI’s side.