Anthropic's New AI Is Its Smartest Yet, and Cheaper to Use
Anthropic released Claude Opus 5.5 on September 22, 2026. Artificial Analysis lists its Adaptive Reasoning, Max Effort, Default Fallback version with a test score of 58, ranked No. 1 out of 206 models. Artificial Analysis
The listing names Anthropic as the maker and gives the same release date. It says the model takes in text and images, writes text answers, and can reason step by step. It can look at up to 1M tokens at once, like a very large short-term memory. Artificial Analysis Anthropic dates its system prompts for Claude Opus 5.5 to September 22, 2026. Anthropic
The price list is $4.00 per 1M input tokens and $20.00 per 1M output tokens, with a 95% discount when the model reuses recent input. Tokens are small pieces of words the model reads and writes. Artificial Analysis For comparison, Claude Opus 5 cost $5 per million input tokens and $25 per million output tokens, the same as Claude Opus 4.8. Anthropic
Anthropic says Opus 5.5 costs 40% less to run than Opus 5. The price list alone points to a 20% drop. The gap matters because the full cost also depends on processing speed, reuse of past input, and how much thinking the model does per question.
Artificial Analysis puts the average cost per test task at $5.98 for the tested setup. Artificial Analysis The full test produced 260M output tokens and cost $8708.20 to run. Artificial Analysis
Anthropic says Opus 5.5 performs at the level of Claude Fable 5.1 on most work. It described the earlier Opus 5 in more careful terms, as a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5. Anthropic It also called Opus 5 its strongest Opus model yet for running long, multi-step helper software, with better coding and professional work.
Anthropic called Claude Opus 4.5 its flagship model with top coding and helper performance at lower pricing. Anthropic It announced Opus 5 as available on all platforms at the same pricing as Opus 4.8. Anthropic Its system card called Opus 5 its latest large language model and an upgrade to Opus 4.8. It says Claude Sonnet 5's early pricing of $2 per million input tokens and $10 per million output tokens is now permanent. Anthropic
One outside data point was included. Sacks said GLM-5.2 is "just a tick below Anthropic's Opus 4.8 and right up there with OpenAI's GPT 5.5." Reuters
The broader context here is price for what you get, not rank alone. A No. 1 score with the thinking turned all the way up shows what is possible when cost is no object. It does not show the cost with normal settings, heavy reuse, or long documents where input text dominates.
In my view, the long name helps. Adaptive Reasoning, Max Effort, Default Fallback names three choices users face: how much thinking to allow, how hard to push, and what happens when limits hit. I have seen this with my own children. They try the smartest mode once, then stick with the cheaper default. Many businesses do the same after the bill arrives. That makes the $5.98 per task as telling as the 58, and points to what becomes possible if the price keeps falling while the smarts stay.
