According to CoinMeta, Claude successfully designed 14 out of 15 targets in a round of protein design experiments. Researchers equipped Claude with a set of protein design and structure prediction tools, enabling it to independently search for information, select design locations, call upon these tools, and continuously filter and modify plans. Ultimately, through real biological experiments conducted by two biotechnology companies, Adaptyv Bio and Twist Bioscience, 354 out of 1320 designs were successful, resulting in a success rate of about 27%. Anthropic noted that such protein designs typically have a success rate of only around 10% to 15%, and some targets even surpassed the winning solutions from previous protein design competitions. Nevertheless, Claude still has a long way to go before developing a new drug; designing a protein that can bind to the target is just an early step in drug development, and numerous further experiments and clinical trials are required thereafter.