Anthropic has released Claude 3, the third generation of its AI model. The release consists of three versions — Haiku, Sonnet, and Opus — each targeting different use cases and price points. Claude 3 can summarize texts of up to 150,000 words and is said to outperform competing language models in several benchmarks.
Performance benchmarks
According to Anthropic, the Claude 3 Opus model outperformed GPT-4 and Gemini 1.0 Ultra from competitors OpenAI and Google in tests covering college-level reasoning, programming, and mathematics. Opus is the most advanced and most expensive model in the lineup. Haiku is the smallest version, designed to be the fastest and most cost-effective option, capable of reading a compact research report with diagrams and graphs in under three seconds.
What this means for Fozzels
At Fozzels, we will be adding more large language models (LLMs) to our service in addition to OpenAI's ChatGPT, so that our customers can choose which AI service provider to use for automatic batch content generation. Follow us on LinkedIn for updates on when Claude 3 becomes available in Fozzels.
Multimodal support and expanded context
This is the first time Claude offers multimodal support. Claude 3 can handle both text and image input and should be able to summarize texts of up to 150,000 words — twice as many as its predecessor. According to Anthropic, the third generation can also understand more context, process longer instructions, and be more accurate than Claude 2.1. Sonnet, the middle model, is reportedly twice as fast as the previous generation and excels at tasks that require quick responses.
Training and infrastructure
Anthropic's Claude 3 models were trained on both publicly available information as of August 2023 and non-public third-party data. Hardware from Amazon Web Services and Google Cloud was used to train the language models. Both Amazon and Google have invested in the startup.
Pricing and availability
Opus and Sonnet were made available immediately, with the release of Haiku following shortly after. Anthropic charges $15 for every million tokens processed by Claude 3 Opus. By comparison, OpenAI charges $10 for every million tokens processed by GPT-4 Turbo. For Sonnet and Haiku, pricing is set at $3 and $0.25 per million tokens, respectively. A limited free version of Claude is also available.
