Analysis: GPT-6 Astra and Claude Fable 5.1 bill 2x apart at equal list price

On October 9, 2026, DEV Community published an analysis by Kamil Kępiński, founder of SmophyAI, comparing the real-world cost of GPT-6 Astra and Claude Fable 5.1. According to the piece, the two models carry identical list prices: $10 per million input tokens and $50 per million output tokens. Yet actual bills can differ by up to 2x in either direction, depending on the workload.

The analysis highlights three points it says the pricing pages do not emphasize. First, Astra's cached input costs $1.00 per million tokens, four times Claude's $0.25 per million. Second, once input exceeds 272K tokens, Astra's rates double to $20 per million for input and $75 per million for output. Third, independent measurements reportedly show Astra completing benchmarks differently from Claude, though the excerpt cuts off before giving specific figures.

The report does not describe the testing methodology, sample size, or time frame behind the comparison, and it does not include any response from either company. The analysis currently stops at the conclusion that identical list prices can translate into roughly 2x different bills, with no further developments reported.