Anthropic is a US AI company founded by former members of OpenAI.

Claude 3.5

In June 2024, Anthropic announced an updated Claude Sonnet model—Claude Sonnet 3.5. As with all the upgrades, the new model is more intelligent, faster, and less expensive than the previous versions.

Two notable features make Claude Sonnet 3.5 a strong contender in the fast-evolving LLM land. Claude 3 already had the best vision ability when analysing charts, graphs and slides; now Claude 3.5 Sonnet claims to be even better.

Another power feature, artifacts, allows users to generate code, presentations, web design, and code and preview them in dedicated windows (environments) alongside the chat. This allows users to see, edit, and build upon the outputs of Claude models. This type of interaction enables a seamless copilot experience when working with AI, removing the reliance on developing custom agents and tools for complex tasks.

Claude 3

Claude 3 is a family of LLMs comprising three multimode models (language and vision): Opus, Sonnet, and Haiku, ranging from the most powerful to the least expensive.

Upon release, Claude 3 models quickly topped various benchmarking leaderboards in language processing, maths, reasoning, common knowledge, and code generation, entering the league of top-performing Gen AI models and making it a formidable opponent to GPT-4 and Gemini Pro 1.0.

A few features stand Claude 3 out:

  • Opus can handle up to 1 million tokens (upon request)
  • Sonnet and Haiku achieved top marks in AI2D and chart-based Q&A
  • Widely touted [@armstrongClaudeMostHuman2024] to have the most human-like responses and is above Gemini Pro on the LMSYS Chatbot Arena Leaderboard
  • Opus achieved the top mark in the “needle-in-the-haystack” test
  • Its excellent next token prediction resulted in itself declaring self-aware

Claud 3 is available as API with tools support on Claude API, Amazon Bedrock and a private preview on Google Cloud. Anthropic also provides a ChatGPT-like chatbot named Claude Pro powered by Sonnet, which costs the same as ChatGPT’s $20/month to use the top-tier Opus model.

References