Anthropic, a startup supported by Google and Amazon, unveiled its newest creation, Claude 3, on Monday.
The model from this new suite, Claude 3 Opus, is reported to surpass the capabilities of its contemporaries, including OpenAI’s GPT-4 and Google’s Gemini 1.0 Ultra, across a range of standardized tests.
The tests were particularly done through benchmark tests, such as undergraduate level knowledge, graduate level reasoning, and basic mathematics. According to CNBC, Anthropic has not offered multimodal support until now. From this viewpoint, users have the capability to submit images, graphs, documents, and various forms of unstructured data to be analyzed and to receive responses based on that analysis.
Claude 3's retrieval ability over long content is so good that if you provide structured data, it essentially acts like a fine-tune! 🤯
I gave it 800 of my tweets and now it just speaks like me. It's hard to convey my level of shock…
It even tagged @Replit and @everartai! pic.twitter.com/sN6zOVBdH8
— Pietro Schirano (@skirano) March 6, 2024
Daniela Amodei, the President of Anthropic, highlighted that despite its higher cost, customers would prefer Claude 3 Opus for its ability to adeptly manage highly complex cognitive tasks, such as conducting precise financial analyses. Anthropic has set the price for Claude 3 Opus at $15 per million tokens processed, which is the unit of data the model uses.
This pricing is significantly higher compared to its less advanced models, which are offered at a cost that is at least five times lower for processing the same amount of data. In comparison, OpenAI’s pricing for the GPT-4 Turbo model is $10 per million tokens, making Claude 3 Opus a more premium option.
Sonnet and Haiku
Sonnet and Haiku, the other models, are characterized by their more compact design and affordability compared to Opus. Sonnet and Opus would be accessible in 159 countries, as announced by Anthropic, with Haiku expected to be released shortly.
The duration and financial investment required to train Claude 3 were not disclosed by the company. However, Anthropic acknowledged that corporations such as Airtable and Asana contributed to the A/B testing of the models.
Anthropic has stated that Sonnet, an intermediary model, delivers performance at twice the speed of both Claude 2 and Claude 2.1. This efficiency, according to the firm, makes it particularly adept at handling tasks that require quick turnarounds, such as retrieving information or automating sales processes. Claude 3 Opus, Sonnet, and Haiku are the names of the new tools, but Claude 3 was dubbed the most powerful.
Omg. I'm sold. I love Claude now. And this is just Claude 3 Sonnet. Imagine Opus pic.twitter.com/4TAN0awYNm
— Actual Genius Intellectual (@ActualGeniusInt) March 7, 2024
Training Claude 3 models
As of August 2023, Anthropic developed the Claude 3 models utilizing a diverse combination of proprietary, third-party, and publicly accessible datasets. In a document detailing the introduction of the three models, Anthropic disclosed that the training process leveraged computing resources from Amazon Web Services (AWS) and Google Cloud.
Notably, both Amazon and Google have made investments in Anthropic, with Amazon contributing $4 billion to the enterprise. Claude 3 is set to be accessible through AWS’s Bedrock model library and Google’s Vertex AI platform.
Anthropic recently disclosed that their latest AI, Claude 3, has enhanced capabilities, notably in its ability to process and summarize extensive texts up to approximately 150,000 words. This capacity aligns with the volume of substantial literary works, comparable to the length of “Moby Dick” or “Harry Potter and the Deathly Hallows,” doubling the summarization ability of its predecessor, which was limited to 75,000 words. Claude 3 allows users to feed in large datasets and request summaries tailored as memos, letters, or stories, whereas ChatGPT’s capacity is around 3,000 words.
and then