Claude 2

Definition

Claude 2 was Anthropic's second-generation large language model, released in July 2023, offering a 100K-token context window and improved reasoning, coding, and safety behavior over Claude 1.

Overview

Claude 2 launched in July 2023 as Anthropic's first Claude model made broadly available to consumers through a public claude.ai website (initially in the US and UK), alongside continued API access for businesses. It followed Claude 1 and Claude Instant, and represented a significant jump in benchmark performance, particularly on coding tasks (scoring notably higher on the Codex HumanEval benchmark) and standardized tests such as the Bar exam and GRE, which Anthropic highlighted in its release materials. A defining feature of Claude 2 was its 100,000-token context window — large enough to process roughly 75,000 words, or several hundred pages of text, in a single prompt. This made it well-suited to summarizing long documents, analyzing contracts, and answering questions over large codebases, use cases that were difficult with the shorter context windows common at the time. Claude 2 was trained using Anthropic's Constitutional AI approach, in which the model critiques and revises its own outputs against a set of guiding principles, combined with reinforcement learning from human feedback, aimed at making responses more helpful and less likely to produce harmful or toxic content. Anthropic later shipped an incremental update, Claude 2.1, in November 2023, which doubled the context window to 200K tokens and reduced hallucination rates. Claude 2 was eventually superseded by the Claude 3 family (Haiku, Sonnet, Opus) in March 2024, and subsequently by Claude 3.5 and later generations, after which Claude 2 was deprecated from active API offerings. It remains notable as the model generation that established Claude as a mainstream competitor to GPT-3.5/GPT-4 and marked Anthropic's shift from an API-only research lab to a consumer-facing AI company.

Key Features

• 100,000-token context window at launch, expanded to 200K in the Claude 2.1 update
• Trained with Constitutional AI and reinforcement learning from human feedback
• Notable improvements in coding benchmark performance over Claude 1
• First Claude model offered directly to consumers via claude.ai
• Available through the Anthropic API for enterprise and developer use
• Reduced hallucination rate compared to Claude 1, further improved in 2.1
• Supported document upload and long-form analysis in the chat interface

Use Cases

Alternatives

Frequently Asked Questions

Related in AI Models