Claude (Anthropic models)
Anthropic's LLM family. Three-tier naming: Haiku (fast + cheap), Sonnet (balanced), Opus (most capable). Versions: Claude 2 (2023), Claude 3 (March 2024), Claude 3.5/3.7 (2024), Claude 4 + 4.5 (2025).
Differentiating: Constitutional AI training (RLAIF — AI as judge), longer thinking modes (Extended Thinking since 4.5), industry-leading code benchmarks (SWE-bench). API + Claude.ai chat. Claude Code (CLI agent), Computer Use (desktop control). Pricing tier 2025: Haiku ~$1/M, Sonnet ~$3/M, Opus ~$15/M input. Competitor: GPT-5 (OpenAI), Gemini 2.5 (Google). Claude excels on agent tasks, code generation, long context (200k stable, 1M experimental). Safety narrative: differential from OpenAI's more product focus.