Anthropic and OpenAI list equal lower-tier prices for Claude Haiku 5.5 and GPT-6 Luna, but Haiku 5.5 shifts to a higher schedule above 100,000 input tokens, making prompt length central to the comparison.
As of Nov. 2, the lower tier for both models sets input at $0.10 and output at $0.50 per million tokens, for requests with up to 100,000 input tokens. For Haiku 5.5 requests beyond that threshold, Anthropic's higher tier lists output at $2.50 and input at $0.50 per million tokens.
OpenAI's GPT-6 Luna retains its lower schedule for prompts containing up to 272,000 input tokens. For Haiku 5.5, the higher schedule covers the whole request once input crosses 100,000; it is not a surcharge confined to the extra tokens. The schedules therefore align only across a limited prompt range.
Anthropic selects a tier by the input length of each request. Prompts no longer than 100,000 tokens use the lower schedule; requests above that mark move to higher prices. The change applies to the full request, not just the excess tokens.
The final bill varies with the request’s input and generated output token counts. The listed prices apply when customers run the models; training is a separate expense.
From 2016, estimated hardware and energy expenses for final training of 45 frontier models rose about 2.4-fold annually. Another estimate, based on five benchmarks, found inference costs falling by roughly 47% per quarter since 2023; it compared the least costly models that cleared performance thresholds. That benchmark trend does not establish an equivalent decline in every API price or real-world task, nor whether inference savings offset training and infrastructure expenses.
Anthropic announced Haiku 5.5 on Oct. 7, saying the model was built for rapid responses across large volumes of requests. EconoTimes previously covered a user's claim that an unreleased Anthropic model might infer intent from brief prompts.
