Claude: definition + examples

Claude is a series of large language models (LLMs) created by Anthropic, a public-benefit AI company founded in 2021 by former OpenAI researchers including Dario Amodei and Daniela Amodei. The Claude family includes Claude Instant (a faster, cheaper variant), Claude 2, Claude 3 (Haiku, Sonnet, Opus), Claude 3.5 (Sonnet, Haiku), and Claude 4 (Sonnet, Opus) as of early 2026. The models are built on transformer architectures with billions of parameters—Claude 3 Opus reportedly uses over 2 trillion parameters in a mixture-of-experts (MoE) configuration, though Anthropic has not disclosed exact sizes. Claude is trained using a combination of supervised fine-tuning, reinforcement learning from human feedback (RLHF), and a proprietary technique called constitutional AI (CAI). CAF trains the model to follow a set of written principles (a “constitution”) that guide its behavior toward harmlessness and helpfulness without relying solely on human raters. This reduces the need for extensive human oversight and aims to produce more predictable, aligned outputs. Claude is accessed primarily via API (Anthropic API), a web interface (claude.ai), and enterprise integrations. It supports a context window of 200,000 tokens (Claude 3.5/4 models), enabling long-document analysis, code generation, and multi-turn conversations. Claude excels at nuanced reasoning, creative writing, and tasks requiring careful instruction-following. It competes directly with OpenAI’s GPT-4, Google’s Gemini, and Meta’s Llama models. Common use cases include legal document summarization, customer support chatbots, medical literature review, and software development assistance. A notable pitfall is that Claude can be overly cautious, refusing harmless requests due to safety filters, especially in earlier versions. Its refusal rate has decreased with later models (e.g., Claude 4 Opus shows 30% fewer unnecessary refusals than Claude 3). As of 2026, Claude 4 Opus is considered state-of-the-art for alignment and safety benchmarks, scoring 92% on the Anthropic Safety Benchmark and 89% on MMLU (massive multitask language understanding). Anthropic also publishes safety research and model cards, contributing to transparency in AI development.