Anthropic's updated tokenizer, shipped with Sonnet 5 and Opus 4.7/4.8, causes the same input to map to 1.0-1.35x more tokens, leading to higher effective costs for users. Playcode's analysis found Claude's new tokenizer consumes up to 73 percent more tokens for a TypeScript file compared to OpenAI's GPT-5.x models. This token inflation could raise Anthropic's effective input costs to $7.50/M and output costs to $37.50/M, despite introductory discounts.