Opus 5.5: Fable 5.1 level, 30% faster and 40% cheaper than Opus 5
Anthropic releases Opus 5.5 this week. Anthropic says that it performs at a level similar to Fable 5.1. It is also 30% faster and 40% cheaper than Opus 5. Claude Code increases the 5-hour window limit by 20% and resets the limits so that you can try the model.
The benchmarks show that Opus 5.5 at medium effort is better than Fable 5.1 at high effort. Rafa's opinion is more careful: Opus 5.5 at medium is equal to Fable 5.1 at low. But it is two times faster and it uses fewer tokens for the same task. Like Fable 5.1, Opus 5.5 lets you change the effort in the middle of a task and keep the cache.
Addy Osmani publishes a guide on the official Claude blog. The main advice: define clearly what a finished task is. For example, the code runs in production, all the tests pass, or the design matches the Figma file. With this, you do not need /goal or prompts like "think harder". Rafa runs 3-hour sessions and uses less than half of his 5-hour window.
Rafa was moving away from Claude Code because Opus 5 gave very verbose output. With Opus 5.5, he uses Claude Code all the time again. Javi adds data from Has Been Released: the price per token is lower than Opus 5, but the price per task is higher. The OpenAI conference is next week, so the competition continues. Watch in the video.

All the news we covered