Everyone is waiting for Opus 5.2, but the latest reports show that the next version number has jumped directly to 5.5. Early today, developer Lyra revealed a hot tip: Claude Opus 5.5, codenamed "claude-wafer-eap," is already being secretly tested, and it's rumored that if you act quickly, it might be released suddenly on Tuesday.
Pricing Rumors: Input $4, Output $20
Evidently, Opus 5.5 is aiming directly at GPT-6. Even more surprising is the pricing list that was exposed: just $4 per million Tokens input and $20 per million Tokens output, a reduction of about 20% compared to the current Opus 5's $5 and $25.
The key lies in the significant optimization of the caching mechanism—caching read costs have dropped to $0.2 per million Tokens, and writing to cache is $5. A $0.2 per million Token caching read rate means the cost for enterprises building long-term, long-context autonomous agent systems will drop by an order of magnitude. With this pricing strategy, Anthropic is turning top models from "luxury items" into "daily necessities."
The release schedule also seems unusual: Claude Opus 5 was officially released on July 24th, followed by a brief test checkpoint for 5.2, and now it's jumping straight to 5.5. Developer Harshith participating in the internal testing said, "Opus 5.5 is a big win, much better than previous Fable 5.1 and Astra," and had 5.5 generate a 3D model of a Waymo self-driving car, with clear details like the Jaguar logo and top lidar sensor, at half the cost of Opus 5. Another developer created a web-based 3D car viewer using Three.js and a "Walle robot" control console, with UI layout and mechanical integration quite solid.
Fable also has new updates: developer Pankaj Kumar released a preview experience for Fable 5.2, saying that the official seems to be quietly switching from 5.1 to 5.2. He rated its 3D generation capabilities as having caught up or even surpassed Astra—without detailed prompts, it can accurately understand requirements and generate a "cliffside villa" with detailed lighting, material reflection, and atmosphere. It can also directly generate a complete game using Three.js. Another developer tested it without providing materials or plugins, and with just native JS, created a "mini world" with dozens of top modeling elements, facial micro-expressions, dynamic effects, and interaction logic all generated at once and running smoothly.
