Anthropic Releases Claude 4.5 Sonnet
Anthropic has released Claude 4.5 Sonnet, positioning it as a 'balanced' model for text and image tasks. It's designed for strong performance with good speed and cost efficiency, featuring a 999,000-token context window. The model is aimed at startups needing a pragmatic choice for production workflows that balances capability with cost.
Anthropic's latest release is actually Claude 3.5 Sonnet, the first in its new 3.5 model family. It outperforms their previous top-tier model, Claude 3 Opus, on a wide range of evaluations while being twice as fast and significantly cheaper to run. The company plans to release Claude 3.5 Haiku and Claude 3.5 Opus later this year to complete the family. On benchmarks, Claude 3.5 Sonnet sets new standards for graduate-level reasoning (GPQA), undergraduate knowledge (MMLU), and coding proficiency (HumanEval). In an internal agentic coding evaluation, it solved 64% of problems, a marked improvement over the 38% solved by Claude 3 Opus. Its vision capabilities are also state-of-the-art, surpassing Opus on standard vision benchmarks, with notable improvements in interpreting charts and transcribing text from imperfect images. The model is priced at $3 per million input tokens and $15 per million output tokens, accessible via the Anthropic API, Amazon Bedrock, and Google Cloud's Vertex AI. It features a 200,000 token context window, which is larger than GPT-4o's 128,000 tokens, allowing it to process and analyze extensive documents or codebases. A new feature called "Artifacts" has been introduced on Claude.ai, creating a dynamic workspace where users can edit and build on the model's creations in real-time. This transforms the tool from a conversational AI into a collaborative work environment. Additionally, a "computer use" capability is in public beta, allowing the model to perceive and interact with computer interfaces by using a cursor, clicking, and typing. Startups are increasingly adopting Anthropic's models. Within the Y Combinator accelerator, Claude has overtaken OpenAI as the most popular model choice for its latest batch of companies, with over 52% of firms preferring it. Companies like HumanLayer and Vulcan have built their entire platforms using Claude, enabling even non-technical founders to ship products. Despite the performance leap, Anthropic classifies Claude 3.5 Sonnet at the same safety level (ASL-2) as its predecessor. The company, which has major offices in San Francisco, continues to hire for a wide range of roles, from research and engineering to policy and communications, emphasizing its mission to build safe and beneficial AI.
