Anthropic Model Changelog

The Evolution of Claude

Anthropic has released multiple generations of Claude models since the company was founded in 2021. Each generation has brought significant improvements in reasoning ability, context length, multimodal capabilities, and developer tooling. This changelog tracks every major release, from the original Claude through the latest Claude 4 family, providing a comprehensive reference for developers who need to understand what changed between versions and how to migrate their applications.

The model naming convention follows a pattern of generation number and tier. Each generation includes up to three tiers: Opus (most capable, highest cost), Sonnet (balanced performance and cost), and Haiku (fastest, most affordable). Not every generation released all three tiers simultaneously. The model identifiers used in API calls include the tier name, generation, and a date suffix that identifies the specific release, such as claude-sonnet-4-20250514.

Generation 1: The Foundation

Claude 1, released in March 2023, was Anthropic's first publicly available model. It supported a 9,000-token context window and demonstrated strong performance on conversational tasks. Claude 1.3 (also known as Claude Instant) was a faster, more affordable variant optimized for high-throughput applications. These early models established the foundation for Anthropic's approach to AI safety with Constitutional AI (CAI) training.

The original API was simpler than today's Messages API, using a text completion format. There was no vision support, no tool use, and no streaming. Developers interacted with Claude through a single prompt string. Despite these limitations, Claude 1 quickly gained adoption for its strong instruction following and lower hallucination rates compared to competitors available at the time.

Generation 2: The 100K Breakthrough

Claude 2, released in July 2023, was a transformative upgrade. The context window expanded from 9K to 100K tokens, enabling entire books, codebases, and document collections to be processed in a single request. This was the largest context window available from any major AI provider at the time and opened new use cases in document analysis, code review, and long-form content generation.

Claude 2.1, released in November 2023, improved accuracy and reduced hallucination rates. It introduced a 200K token context window in beta, system prompts for controlling Claude's behavior, and tool use (function calling) as an experimental feature. These additions transformed Claude from a conversational model into a platform that developers could build complex applications on. Claude 2.1 also significantly improved coding ability and mathematical reasoning.

Generation 3: The Multimodal Era

Claude 3, released in March 2024, was the most significant update in Claude's history. It introduced the three-tier model family (Opus, Sonnet, Haiku), vision capabilities for all tiers, a standard 200K context window, and dramatically improved benchmark performance across all categories. Claude 3 Opus set new records on graduate-level reasoning benchmarks, while Claude 3 Haiku provided unprecedented speed at a fraction of the cost.

Claude 3.5 Sonnet, released in June 2024, was a landmark mid-generation release that matched or exceeded Claude 3 Opus performance on most benchmarks while maintaining Sonnet-tier pricing. This effectively gave developers Opus-level quality at Sonnet-level cost. Claude 3.5 Sonnet became the default recommendation for most production applications and demonstrated that mid-generation updates could deliver generational improvements. Claude 3.5 Haiku followed with similar efficiency gains at the Haiku tier. For comparing model benchmarks across providers, LockML provides cross-provider comparison tools.

Generation 4: Current State

Claude 4, released in May 2025, represents the current generation with Claude 4 Opus and Claude 4 Sonnet. These models bring further improvements in coding, reasoning, instruction following, and multilingual capabilities. Claude 4 Opus is the most capable model Anthropic has released, with top scores on advanced reasoning and code generation benchmarks. Claude 4 Sonnet continues the pattern of delivering excellent quality-to-cost ratio for production applications.

The API ecosystem around Claude 4 includes prompt caching (reducing costs by up to 90% for repeated content), message batches (50% cost reduction for async workloads), extended thinking (step-by-step reasoning for complex problems), and improved tool use reliability. These features make Claude 4 models suitable for enterprise-scale applications with complex workflows. For tracking costs across model versions, KickLLM provides model pricing calculators.

Key Capability Milestones

Vision was introduced with Claude 3 in March 2024, supporting JPEG, PNG, GIF, and WebP images up to 20MB. All subsequent models maintain vision support. Tool use was introduced experimentally in Claude 2.1 and became production-ready in Claude 3. Streaming was available from Claude 2 onwards. Prompt caching was introduced as a platform feature available to Claude 3 and later models. Message batches launched in 2024 for high-volume processing at reduced cost.

Extended thinking was introduced with Claude 3.5 Sonnet and enhanced in Claude 4, allowing Claude to show its reasoning process step by step. Computer use was introduced as a beta feature allowing Claude to interact with desktop environments through screenshots and mouse/keyboard actions. These capabilities continue to evolve with each model release, making Claude increasingly versatile as a development platform.