According to CoinMeta, Tibo reminds us that when comparing the prices of different AI models, we cannot simply look at “how much it costs per million token.” In GPT-5.6 and SOL, 766 token are used, while Claude Opus estimates that 1170 are needed, which is about 34.5% less for SOL. Currently, both models have a standard input price of $5 per million token, with an output cost of $30 for SOL and $25 for Opus. If we only consider the output price per unit, Opus seems cheaper. However, when calculating based on output, SOL costs approximately $0.023, Opus costs around $0.029, and SOL is actually about 21% cheaper. To use pizza as an analogy, a more reasonable way to compare is to see “how much it actually costs to complete a task,” and we also need to take into account the model’s success rate and the actual amount of token used.