Anthropic's New Claude Does the Same Work for Less Money

Anthropic announced Claude Opus 5.5 on Sep 22, 2026. On the same day, sample code in the anthropics/ClaudeForFoundationModels project changed from ClaudeLanguageModel(name: .opus5) to ClaudeLanguageModel(name: .opus5_5). Commit

The company says Opus 5.5 does most work at the level of Claude Fable 5.1. It also says Opus 5.5 costs 40% less to run than Opus 5. Anthropic Newsroom

Opus 5 had been described as near Claude Fable 5 at half the price. It was presented as good for daily office work and computer programming. Reuters Opus 5.5 keeps that trend, with Fable 5.1-level results on most work at a lower run cost than Opus 5.

For programmers, the Sep 22 note says .opus5_5 stands for the model name claude-opus-5-5. That is the exact name typed to pick the model. The edit is one name swap, but that name is used for routing, logging, and access control.

The same note explains how names work. Dateless names like claude-opus-5-5, from version 4.6 onward, are frozen copies. They do not auto-update. Think of a printed photo, not a live video feed. The name points to the same fixed version. Teams that need steady results for testing, retesting after changes, or checked and approved use can count on it staying the same. Teams that want improvements must choose a new name.

For background, Anthropic calls Claude Opus 5 thoughtful and proactive, close to the top smarts of Claude Fable 5. Anthropic Opus 5 Announcement It lists May 2026 as the cutoff date for what Opus 5 was taught. Anthropic Transparency Separately, Anthropic said Claude now leads 26% of AI research and development work inside the company. Reuters

The bigger picture is about cost and trust. Budgets often decide if a capable model stays as a small test or becomes an everyday tool for coding help, writing drafts, and automated tasks. A 40% lower run cost changes planning without needing a separate smaller model for routine calls. Frozen copies lower the risk of surprise shifts in tuned prompts, and they make side-by-side tests cleaner across answer speed, tool use, and output quality.

In my view, the frozen-copy rule may last as long as the price cut. I have watched my own children pick up new tools fast and drop them just as fast when the behavior shifted without warning. Business users act the same way on a larger scale. Predictability builds trust. Platform teams will need to add claude-opus-5-5 and .opus5_5 to approved lists, cost trackers, and testing tools, while keeping older names for backup and comparison. That should make it easier to use a strong helper every day.