Claude Sonnet 5 continues Anthropic's pattern of hiding price increases behind unchanged token rates

The more pointed issue isn't that Sonnet 5 costs more in practice, it's that Anthropic has now done this across multiple model generations, suggesting a deliberate pricing architecture rather than an incidental efficiency trade-off. Buyers who benchmark on list price alone are being systematically mispriced.

This story lands on the same day The Decoder reported that OpenAI is fragmenting GPT-5.6 Pro into three distinct variants, a move that also complicates apples-to-apples cost comparisons across the frontier. Both developments point toward a market where nominal pricing is increasingly a poor proxy for actual deployment cost, and where buyers need production telemetry rather than published rate cards to make sound procurement decisions. The Fable 5 reinstatement coverage from TechCrunch and The Verge this week adds a separate wrinkle: Anthropic is simultaneously managing regulatory friction, safety incidents, and now a pricing opacity story, which collectively raise questions about enterprise trust at a moment when the company needs to consolidate commercial momentum.

Watch whether enterprise benchmark aggregators like Artificial Analysis update their cost-per-task methodology to weight token consumption by model within the next two quarters. If they do, Anthropic's effective price premium will become visible in standard comparisons and force a public response on efficiency.