Claude Opus 4.7: Anthropic raises the bar and shifts the price unit

Following a leak by The Information on April 14th and a week of speculation on Reddit, Polymarket, and in the usual AI newsletters, the model is now live: on the Claude platform, via the API, and on all three major cloud providers—Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Model ID for developers: claude-opus-4-7.

Two things make this release more interesting than another point-seven iteration update.

1. According to VentureBeat, Opus 4.7 is currently the most powerful LLM available generally — just ahead of GPT-5.4 and Gemini 3.1 Pro. It leads quite clearly in the benchmarks that matter to professional users (coding, vision, finance).
2. And that's the more honest story—Anthropic itself admits in its release communications that Opus 4.7 isn't its most powerful model. That position still belongs to a system called Mythos Preview, which only about forty select companies are allowed to use through the Project Glasswing program. A frontier lab that publicly says, "You're not getting our best model." That's become rare in this industry in 2026.

And then there's the tokenizer issue. More on that later.