Reference

Anthropic Claude Models

Every Claude model ever released by Anthropic, from the original Claude to the latest Claude Opus 4 and Sonnet 4. All models feature Anthropic's Constitutional AI safety approach.

The Constitutional AI Approach

All Claude models are trained using Constitutional AI (CAI), Anthropic's distinctive training methodology. Instead of relying solely on human feedback to define harmful outputs, CAI gives the model a set of principles (a "constitution") that it uses to critique and revise its own outputs during training. This produces models that are helpful yet resistant to misuse, with transparent reasoning about safety decisions.

Current Generation: Claude 4 Family

The latest Claude models, released in mid-2025, represent the most capable generation yet.

ModelReleasedContextMax OutputMultimodalPricing (Input/Output per 1M)
Claude Opus 4May 2025200K32,000Text, Vision$15.00 / $75.00
Claude Sonnet 4May 2025200K16,000Text, Vision$3.00 / $15.00

Claude Opus 4

Anthropic's most intelligent model. Claude Opus 4 is designed for the most complex tasks requiring deep analysis, nuanced understanding, and extended reasoning. It excels at research, complex coding, and tasks requiring careful judgment.

  • Best for: Complex research and analysis, agentic coding, extended reasoning tasks, high-stakes decisions
  • Key features: Extended thinking, tool use, computer use, superior instruction following, 200K context
  • Standout capabilities: Best-in-class for agentic tasks, sustained performance on long tasks, strong at self-correction

Claude Sonnet 4

The ideal balance of intelligence and speed. Claude Sonnet 4 provides near-Opus quality at a fraction of the cost and with faster response times, making it the workhorse model for most applications.

  • Best for: Day-to-day coding assistance, content generation, data analysis, customer-facing applications
  • Key features: Excellent coding, strong reasoning, fast responses, tool use, computer use
  • Standout capabilities: Excellent cost-to-performance ratio, strong at following nuanced instructions

Claude 3.5 Generation

ModelReleasedContextMax OutputMultimodalPricing (Input/Output per 1M)
Claude 3.5 SonnetJun 2024 (updated Oct 2024)200K8,192Text, Vision$3.00 / $15.00
Claude 3.5 HaikuOct 2024200K8,192Text, Vision$0.80 / $4.00

Claude 3.5 Sonnet

A landmark release that outperformed the original Claude 3 Opus on most benchmarks while being faster and cheaper. The October 2024 update (sometimes called "new Sonnet") further improved coding and tool use capabilities. Introduced the computer use capability.

  • Best for: Coding, analysis, writing, general-purpose tasks
  • Historical significance: Demonstrated that a smaller, well-trained model could surpass a larger one, shifting the focus from scale to training quality

Claude 3.5 Haiku

The fastest model in the 3.5 generation. Designed for high-throughput, latency-sensitive applications. Despite its speed, it performs comparably to Claude 3 Opus on many benchmarks.

  • Best for: High-volume chatbots, real-time applications, classification, quick responses

Claude 3 Generation

ModelReleasedContextMax OutputMultimodalPricing (Input/Output per 1M)
Claude 3 OpusMar 2024200K4,096Text, Vision$15.00 / $75.00
Claude 3 SonnetMar 2024200K4,096Text, Vision$3.00 / $15.00
Claude 3 HaikuMar 2024200K4,096Text, Vision$0.25 / $1.25

Claude 3 Opus

The original flagship of the Claude 3 family. At launch, it was the most capable LLM available for complex tasks. Introduced the three-tier Opus/Sonnet/Haiku naming convention.

  • Best for: Complex analysis, research, nuanced writing (now largely superseded by newer models)

Claude 3 Sonnet

The balanced middle-tier model of the Claude 3 family. Good balance of speed and capability for everyday tasks.

Claude 3 Haiku

The fastest and most compact model in the Claude 3 family. Extremely cost-effective for simple tasks. One of the cheapest frontier-class models available at launch.

  • Best for: Lightweight tasks, high-volume processing, real-time responses

Earlier Generations

Claude 2.1

  • Released: November 2023
  • Context window: 200K tokens (a breakthrough at the time)
  • Key improvement: Reduced hallucination rates, better document analysis, introduced tool use
  • Status: Deprecated

Claude 2

  • Released: July 2023
  • Context window: 100K tokens
  • Key features: First 100K context model, improved coding, better at following complex instructions
  • Status: Deprecated

Claude Instant

  • Released: 2023
  • Context window: 100K tokens
  • Key features: Fast, cost-effective companion to Claude 2. Predecessor to the Haiku tier concept.
  • Status: Deprecated
💡
Model naming convention: Anthropic uses a tier-based naming system: Opus (most capable), Sonnet (balanced), Haiku (fastest). This allows users to quickly identify the right model for their needs without comparing arbitrary version numbers.

Key Differentiators Across All Claude Models

  • 200K context window: Since Claude 2.1, all models support massive context windows, enabling processing of entire codebases, books, or document collections
  • Strong instruction following: Claude models are consistently praised for following complex, multi-constraint instructions precisely
  • Safety by design: Constitutional AI training produces models that are helpful while being resistant to jailbreaks and misuse
  • Honest uncertainty: Claude models reliably express uncertainty rather than confabulating answers
  • Extended thinking: Newer models support a "thinking" mode where the model reasons through problems step by step before responding
  • Tool use: Function calling, computer use, and MCP (Model Context Protocol) support
✅
Which Claude should you use? For new projects: Sonnet 4 for most tasks (best balance), Opus 4 for complex reasoning and agentic workflows, Haiku 3.5 for high-volume, latency-sensitive applications.