Opus 4.7 is smarter, more literal, and quietly more expensive. Those are three different problems.

Source: Nate’s Newsletter Date: 2026-04-21 URL: https://natesnewsletter.substack.com/p/opus-47-is-smarter-more-literal-and

Summary

Nate argues Opus 4.7’s issues are three separate engineering choices, not unified side effects:

1. Increased literalism — the model does exactly what you ask, eliminating inference shortcuts. Where 4.6 interpreted incomplete requests charitably, 4.7 demands precision. Users must reconstruct the interpretive labor the prior version provided “for free.”
2. Hidden cost increases — despite unchanged sticker pricing ($5/$25), actual costs rose via tokenizer changes, adaptive thinking overhead, and breaking API changes. Per-token output costs increased.
3. Non-uniform capability gains — coding, vision, and knowledge improved; web research and terminal functions regressed. Requires strategic routing decisions.

Implications

The pricing analysis connects directly to the token economics thread: the price tag is the same but the bill is higher. Combined with the Copilot token-billing shift and Anthropic’s brief Pro tier removal, this is the third independent data point confirming the subsidy model is breaking. Nate’s “three separate problems” framing is more useful than treating 4.7 as a single upgrade — it means the response has to be three separate adaptations, not one.