Anthropic gives its flagship generative AI models a facelift

The startup company says the latest update to Claude brings more speed, accuracy and an upcoming ‘lite’ version.

LLMs, ChatGPT, Generative AI

One of the hottest startups in the generative AI (genAI) market, Anthropic, has updated its flagship models to a new 3.0 standard, bringing improvements across a range of common tasks and faster processing speeds.

The improvements in Claude 3 are broad-based, according to Anthropic. The model now offers fewer “incorrect” refusals to process harmless requests, better accuracy in its answers, fewer  hallucinations, and better accuracy in processing visual information such as pictures and diagrams.

Anthropic now offers three versions of the Claude AI: the fully-featured Opus, middle-ground Sonnet, and lightweight Haiku. Each version offers different average benchmark scores across various tasks, with the lower-scoring Sonnet and Haiku trading off accuracy for lower costs, in the formeof cheaper tokens for AI calls, and faster response time.

“Haiku is the fastest and most cost-effective model on the market for its intelligence category,” Anthropic said. “It can read an information and data dense research paper on arXiv (~10k tokens) with charts and graphs in less than three seconds. Following launch, we expect to improve performance even further.”

The company was quick to tout its commitment to responsible genAI development, saying it has dedicated teams for a number of particularly serious risks of misuse, including election interference and misinformation, child sexual abuse material, and autonomous replication. In line with guidance issued last year by the Biden Administration, the company said it has performed extensive “red team” exercises to test the potential for malicious use of Claude.

Claude 3 follows closely on the heels of Claude 2.1, which was released last November. That release broadened the model’s context window — the amount of data it can process at once —to 200,000 tokens, or the rough equivalent of half a million written words.

Opus and Sonnet models for Claude 3 are available today via Anthropic’s API, with Sonnet forming the underpinning of the company’s free demo online. Haiku is not yet publicly available, but will be “soon,” according to the company.

Sonnet is also available via Amazon’s Bedrock AI managed service, as well as in private preview on Google Cloud’s Vertex AI Model Garden, with Opus and Haiku on the way to both platforms.

Copyright © 2024 IDG Communications, Inc.

It’s time to break the ChatGPT habit