AI chatbot Claude 2.1 version released, newly added super GPT-4 Turbo, supports 200,000 Token input

On November 23, OpenAI’s competitor Anthropic launched the Claude 2.1 chatbot and corresponding AI model. Claude 2.1 supports the input of up to 200,000 Tokens, improves accuracy, and adds a system preset prompt word (System Prompt) function

The official introduction stated that the previous Claude 2.0 could only support 100,000 Tokens For a single conversation of Token, Claude 2.1 can support the input of up to 200,000 Tokens, which is equivalent to 150,000 texts or 500-page documents, which has exceeded the 128,000 Tokens of OpenAI GPT-4 Turbo (Note from this site: On average, 1 Token corresponds to 4 English characters, and 100 Tokens correspond to approximately 75 English words)

Anthropic said that since the Claude chat service was launched this year, there have been countless Millions of users use it to translate academic reports, draft business plans or analyze complex contracts. Many users want longer tokens when processing large and lengthy documents, so Anthropic decided to double the size of the relevant input.

Official claims that when using the new Claude 2.1, users can enter content or information containing "a book" level for more detailed excerpts, questions and answers, trend analysis, comparisons, etc.

In terms of accuracy, Claude 2.1 claims to be twice as accurate as Claude 2.0. In addition, the model also improves the ability to understand text content, especially for legal documents and financial reports. It has been optimized with higher-accuracy documents such as technical specifications, claiming that "the proportion of incorrect answers has been reduced by 30%."

Claude 2.1 can also make up for possible shortcomings of the general model by integrating domain-specific databases, Anthropic claims, if the user needs For programming, you can ask Claude to switch to "programming mode" in advance to convert natural language into structured output, but this feature is still in the early development stage.

Also, Claude 2.1 now adds the system preset prompt word function that is common among competitors. You can set the prompt words of the model in advance, such as letting the model play a specific role or answer the user's questions in a specific voice

Anthropic also mentioned Claude’s price advantage. Users can use Claude for free to chat, while the price of its paid API is US$8 (approximately RMB 57) per 1 million Tokens for input and US$24 (approximately RMB 172) per 1 million Tokens for output. In comparison, the price of GPT-4 Turbo is higher, with an input of 10 US dollars (about 72 yuan) and an output of 30 US dollars (about 215 yuan) per 1 million Tokens

The above is the detailed content of AI chatbot Claude 2.1 version released, newly added super GPT-4 Turbo, supports 200,000 Token input. For more information, please follow other related articles on the PHP Chinese website!