[AINews] Claude Opus 5.5, the new default model for AINews — and everybody cuts prices 40-50%

Anthropic’s Claude Opus 5.5 landed as the new default in Claude Code and the Claude app, and AINews is moving to it immediately. The pitch is simple enough: it behaves like Claude Fable 5.1 on most tasks, but runs about 40% cheaper than Opus 5. OpenAI answered about an hour later with GPT-6 Sol and Luna, but the day clearly tilted toward Anthropic’s release wave.

The cost story is more complicated than the launch copy makes it sound. The sticker price is down 20%, and Anthropic says the model is about 30% faster and about 40% cheaper per task than Opus 5. But several reports in the source say Opus 5.5 often uses more tokens, especially on coding, so the savings depend a lot on how you run it. At max effort, Artificial Analysis puts it at $5.98 per task, slightly above Opus 5’s $5.86, which is a neat way of saying the cheaper model can still end up costing more if you lean on it hard enough.

What stands out here is not just the benchmark talk. Anthropic made a point of the writing improvements, saying the model puts the important information first and follows writing rules more faithfully. The source says that showed up immediately in AINews tests, and that the difference between Opus 5.5 and Sol 6 was stark enough that AINews is switching over for now. There are also safer defaults, with Opus 5.5 now the default in Claude Code, Claude app, and Cowork, plus a banked rate-limit reset for Pro, Max, and Team users.

On benchmarks, the model looks strong across coding, computer use, and knowledge work, though the source is full of caveats and benchmark drama. Anthropic says it beats Fable 5.1 on every row of its headline comparison and beats GPT-6 Astra on most. Third-party results put it at the top of multiple tables, including Vals Index and CursorBench, while some evaluations warn that higher effort can actually make it worse on certain coding tasks. That weirdness matters. It’s a reminder that these launches are no longer about one clean score line; they’re about which model behaves best under real usage, real costs, and real prompt fatigue.

There’s also a safety angle. Opus 5.5 is the first Opus with Fable 5.1-class safeguards on cyber, bio, and frontier LLM development, and Anthropic says some flagged requests will fall back to another model. The launch thread is trying to sell both speed and restraint at once, which is very on-brand for the current frontier race.