Claude Opus 4.7

Claude Opus 4.7 matters more than a legacy model usually would, because two changes that still shape the whole platform landed with it.

It introduced the current tokenizer

From Opus 4.7 onward, the same text produces roughly 30% more tokens. Every Claude 5 model uses this tokenizer; Sonnet 4.6 and earlier use the previous one.

This is the boundary that makes cross-generation price comparison misleading. A model at $5/MTok on the new tokenizer is not directly comparable to one at $5/MTok on the old one — multiply the older model’s rate by about 1.3 first.

It removed the sampling parameters

temperature, top_p and top_k are deprecated from Claude Opus 4.7 onward. They remain in the SDK request types so existing code type-checks, then return a 400 error at runtime if set to a non-default value.

If an upgrade broke with a 400 and nothing in your types changed, this is why.

Specifications

Property	Value
Claude API ID	claude-opus-4-7
Context window	1M tokens (~555K words on the new tokenizer)
Max output	128K tokens
Adaptive thinking	Yes
Extended thinking parameter	No
Reliable knowledge cutoff	January 2026
Pricing	$5 in / $25 out per MTok

One thing it cannot do

Fast mode is not available on Opus 4.7 — requests with speed: "fast" return an error. It works on Opus 4.8 and Opus 5, and silently no-ops at standard price on Opus 4.6.

Lifecycle

Active but legacy. Tentative retirement not sooner than 16 April 2027.