Claude Opus 4.7

Anthropic's prior flagship, now succeeded by the cost-neutral Opus 4.8 — still capable and available, but new deployments should generally start on 4.8.

Claude Opus 4.7 was Anthropic's top model immediately before Opus 4.8, and it remains available via the API for teams already standardized on it. It's a strong flagship — 1M context, the coding and reasoning strengths Claude is known for — but since 4.8 arrived at the same price with real gains, there's little reason to start a new project here.

One specific thing to know if you're on or considering 4.7: it shipped with a new tokenizer that can generate up to about 35% more tokens for the same input text than Opus 4.6. Per-tokenThe basic unit of text a model reads and writes. Tokens are roughly three-quarters of a word in English — so 100 tokens is about 75 words. Models don't see letters or words directly; they see tokens. Pricing is almost always quoted per million tokens, and context windows are measured in tokens rather than words. prices didn't change, but because you may be billed for more tokens per request, the effective cost can rise — worth benchmarking your own workloads before migrating from 4.6, and a reason the move to 4.8 deserves a quick cost check too.

For most readers this is a stepping-stone entry: useful for understanding the lineage and for teams maintaining existing 4.7 deployments, but Claude Opus 4.8 is the current recommendation.

• Teams already running 4.7 in production who want a stable, supported model.
• Understanding the Opus lineage and the 4.7 tokenizer change before migrating.
• Flagship-tier work where an existing 4.7 integration is already validated.

• New deployments — start on Claude Opus 4.8 (cost-neutral, better).
• Cost-sensitive use sensitive to the tokenizer change that can inflate tokenThe basic unit of text a model reads and writes. Tokens are roughly three-quarters of a word in English — so 100 tokens is about 75 words. Models don't see letters or words directly; they see tokens. Pricing is almost always quoted per million tokens, and context windows are measured in tokens rather than words. counts.
• Everyday work — Claude Sonnet 4.6 is the better-value default.
• Self-hosting — Claude is closed.

Proprietary and API-only, on Anthropic's standard commercial terms: commercial rights to your outputs, none to the model. The notable practical wrinkle is cost, not licensing — 4.7's tokenizer can raise the effective per-request bill versus 4.6 despite identical per-tokenThe basic unit of text a model reads and writes. Tokens are roughly three-quarters of a word in English — so 100 tokens is about 75 words. Models don't see letters or words directly; they see tokens. Pricing is almost always quoted per million tokens, and context windows are measured in tokens rather than words. rates, so benchmark before migrating. No open weightsThe numerical values inside a trained model that encode everything it has learned. A model is, functionally, a giant list of weights — tens of billions of numbers for a mid-sized model, hundreds of billions for a frontier model. "Open-weight" means those numbers are published. "Downloading the weights" means getting the actual file you'd need to run the model yourself. or self-hosting.

Against Claude Opus 4.8, 4.7 is the older model at the same price — 4.8 is the straightforward upgrade. Against Claude Sonnet 4.6, it's the pricier flagship where Sonnet covers most needs. Against GPT-5.5, it competes at the closed flagship tier on coding and reliability, though as a superseded model it's mainly relevant to existing deployments rather than new comparisons.