Model Names

The AnthropicChatModelName enum provides predefined identifiers for Claude models.

Capabilities

Model Name Enum

Predefined Claude model identifiers with version dates.

package dev.langchain4j.model.anthropic;

/**
 * Enum of Claude model identifiers with version dates.
 * Each constant maps to a specific model version string.
 *
 * @since 1.0.0
 */
public enum AnthropicChatModelName {
    /** Claude Opus 4.5 released November 2025 - most powerful with thinking */
    CLAUDE_OPUS_4_5_20251101,

    /** Claude Sonnet 4.5 released September 2025 - balanced intelligence and speed */
    CLAUDE_SONNET_4_5_20250929,

    /** Claude Haiku 4.5 released October 2025 - fastest for simple tasks */
    CLAUDE_HAIKU_4_5_20251001,

    /** Claude Opus 4.1 released August 2025 - previous generation Opus */
    CLAUDE_OPUS_4_1_20250805,

    /** Claude Opus 4 released May 2025 */
    CLAUDE_OPUS_4_20250514,

    /** Claude Sonnet 4 released May 2025 */
    CLAUDE_SONNET_4_20250514,

    /** Claude 3.5 Haiku released October 2024 */
    CLAUDE_3_5_HAIKU_20241022,

    /** Claude 3 Haiku released March 2024 */
    CLAUDE_3_HAIKU_20240307;

    /**
     * Returns API model name string.
     * Format: "claude-{tier}-{version}-{date}"
     * Example: "claude-sonnet-4-5-20250929"
     *
     * @return model name for API requests, never null
     */
    public String toString();
}

Using Model Names

Model names can be used with all chat model builders.

import dev.langchain4j.model.anthropic.AnthropicChatModel;
import dev.langchain4j.model.anthropic.AnthropicChatModelName;

// Using enum constant
AnthropicChatModel model = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(AnthropicChatModelName.CLAUDE_SONNET_4_5_20250929)
    .build();

// Using string directly
AnthropicChatModel model2 = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName("claude-sonnet-4-5-20250929")
    .build();

Error Handling:

• IllegalArgumentException: modelName null or empty (if using string method)
• RuntimeException: Model not available for API key (at runtime during API call)
• RuntimeException: Model name typo (at runtime during API call)

Common Pitfalls:

❌ DON'T use hardcoded strings

.modelName("claude-sonnet")  // Ambiguous, will fail

✅ DO use enum constants or full strings

.modelName(AnthropicChatModelName.CLAUDE_SONNET_4_5_20250929)  // Type-safe
// OR
.modelName("claude-sonnet-4-5-20250929")  // Explicit version

Null Safety:

• Enum constants are never null
• toString() returns never null
• String parameter methods throw if null provided

Model String Values

Each enum constant has a corresponding string value.

String modelString = AnthropicChatModelName.CLAUDE_SONNET_4_5_20250929.toString();
// Returns: "claude-sonnet-4-5-20250929"

Usage Example:

// Get all model names as strings
for (AnthropicChatModelName model : AnthropicChatModelName.values()) {
    System.out.println(model.name() + " -> " + model.toString());
}

// Output:
// CLAUDE_OPUS_4_5_20251101 -> claude-opus-4-5-20251101
// CLAUDE_SONNET_4_5_20250929 -> claude-sonnet-4-5-20250929
// CLAUDE_HAIKU_4_5_20251001 -> claude-haiku-4-5-20251001
// ...

Available Models

Claude 4.5 Series

• CLAUDE_OPUS_4_5_20251101: "claude-opus-4-5-20251101" - Most powerful model with extended thinking

  • Context window: 200K tokens
  • Max output: 16K tokens
  • Supports: Vision, PDF, thinking, tools, caching
  • Best for: Complex reasoning, research, analysis

• CLAUDE_SONNET_4_5_20250929: "claude-sonnet-4-5-20250929" - Balanced intelligence and speed

  • Context window: 200K tokens
  • Max output: 8K tokens
  • Supports: Vision, PDF, tools, caching
  • Best for: General purpose, production workloads

• CLAUDE_HAIKU_4_5_20251001: "claude-haiku-4-5-20251001" - Fastest model for simple tasks

  • Context window: 200K tokens
  • Max output: 4K tokens
  • Supports: Vision, PDF, tools, caching
  • Best for: High-volume, cost-sensitive applications

Claude 4.1 Series

• CLAUDE_OPUS_4_1_20250805: "claude-opus-4-1-20250805" - Previous generation Opus
  • Context window: 200K tokens
  • Max output: 16K tokens
  • Supports: Vision, PDF, tools, caching
  • Best for: Complex tasks without thinking requirement

Claude 4 Series

• CLAUDE_OPUS_4_20250514: "claude-opus-4-20250514" - Claude 4 Opus

  • Context window: 200K tokens
  • Max output: 16K tokens
  • Supports: Vision, PDF, tools, caching
  • Best for: High-quality outputs

• CLAUDE_SONNET_4_20250514: "claude-sonnet-4-20250514" - Claude 4 Sonnet

  • Context window: 200K tokens
  • Max output: 8K tokens
  • Supports: Vision, PDF, tools, caching
  • Best for: Balanced use cases

Claude 3.5 Series

• CLAUDE_3_5_HAIKU_20241022: "claude-3-5-haiku-20241022" - Claude 3.5 Haiku
  • Context window: 200K tokens
  • Max output: 4K tokens
  • Supports: Vision, tools, caching
  • Best for: Fast responses, simple tasks

Claude 3 Series

• CLAUDE_3_HAIKU_20240307: "claude-3-haiku-20240307" - Claude 3 Haiku
  • Context window: 200K tokens
  • Max output: 4K tokens
  • Supports: Vision, tools
  • Best for: Legacy applications

Model Selection Guidelines

Use CLAUDE_OPUS_4_5_20251101 for:

• Complex reasoning tasks
• Extended thinking requirements
• Research and analysis
• Advanced problem-solving
• Mathematical proofs
• Code analysis and debugging

Example:

// Complex research task
AnthropicChatModel model = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(AnthropicChatModelName.CLAUDE_OPUS_4_5_20251101)
    .thinkingType("enabled")
    .thinkingBudgetTokens(10000)
    .build();

Use CLAUDE_SONNET_4_5_20250929 for:

• General-purpose applications
• Balanced performance and cost
• Most production use cases
• Tool use and function calling
• Content generation
• Document analysis

Example:

// Production chatbot
AnthropicChatModel model = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(AnthropicChatModelName.CLAUDE_SONNET_4_5_20250929)
    .temperature(0.7)
    .maxTokens(2048)
    .build();

Use CLAUDE_HAIKU_4_5_20251001 for:

• Simple, fast responses
• High-volume requests
• Cost-sensitive applications
• Quick data transformations
• Classification tasks
• Simple Q&A

Example:

// High-volume classification
AnthropicChatModel model = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(AnthropicChatModelName.CLAUDE_HAIKU_4_5_20251001)
    .maxTokens(512)
    .temperature(0.0)  // Deterministic
    .build();

Usage Examples

Selecting by Use Case

// For complex reasoning
AnthropicChatModel reasoningModel = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(AnthropicChatModelName.CLAUDE_OPUS_4_5_20251101)
    .thinkingType("enabled")
    .thinkingBudgetTokens(10000)
    .build();

// For general use
AnthropicChatModel generalModel = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(AnthropicChatModelName.CLAUDE_SONNET_4_5_20250929)
    .build();

// For fast, simple tasks
AnthropicChatModel fastModel = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(AnthropicChatModelName.CLAUDE_HAIKU_4_5_20251001)
    .maxTokens(512)
    .build();

Dynamic Model Selection

AnthropicChatModelName selectModel(String taskComplexity) {
    return switch (taskComplexity) {
        case "complex" -> AnthropicChatModelName.CLAUDE_OPUS_4_5_20251101;
        case "medium" -> AnthropicChatModelName.CLAUDE_SONNET_4_5_20250929;
        case "simple" -> AnthropicChatModelName.CLAUDE_HAIKU_4_5_20251001;
        default -> AnthropicChatModelName.CLAUDE_SONNET_4_5_20250929;
    };
}

AnthropicChatModel model = AnthropicChatModel.builder()
    .apiKey(apiKey)
    .modelName(selectModel("complex"))
    .build();

Error Handling:

• Invalid taskComplexity falls back to Sonnet (safe default)
• Consider validating input before selection
• Log model selection for debugging

Common Pitfalls:

❌ DON'T always use Opus

.modelName(AnthropicChatModelName.CLAUDE_OPUS_4_5_20251101)  // Expensive for simple tasks

✅ DO match model to task complexity

String complexity = assessComplexity(task);
.modelName(selectModel(complexity))  // Cost-effective

Model Comparison Table

Performance Notes:

• Speed: Relative latency (tokens/second)
• Cost: Relative pricing per 1M tokens
• Context: All models support 200K token context
• Max Output: Configurable via maxTokens(), but cannot exceed model limit

Behavioral Differences Between Models

Intelligence Level

• Opus: Best reasoning, fewer errors, handles ambiguity well
• Sonnet: Good balance, occasional misunderstandings on complex tasks
• Haiku: Best for structured tasks, may struggle with nuance

Response Style

• Opus: Thorough, detailed, considers edge cases
• Sonnet: Concise yet complete, balanced detail level
• Haiku: Direct, brief, focused on task completion

Tool Usage

• Opus: Strategic tool selection, handles complex multi-tool workflows
• Sonnet: Reliable tool use, good for standard workflows
• Haiku: Basic tool use, best for single-tool scenarios

Error Handling

• Opus: Graceful degradation, asks clarifying questions
• Sonnet: Standard error messages, good error recovery
• Haiku: Direct error reporting, minimal recovery attempts

Version Date Significance

Model version dates indicate release snapshots:

• Format: YYYYMMDD (e.g., 20250929 = September 29, 2025)
• Purpose: Explicit versioning for reproducibility
• Updates: New dates released as models improve
• Deprecation: Old versions eventually deprecated (6-12 months notice)

Example:

// Pin to specific version for reproducibility
.modelName("claude-sonnet-4-5-20250929")  // Exact version

// vs. using enum (automatically updated with library)
.modelName(AnthropicChatModelName.CLAUDE_SONNET_4_5_20250929)

Notes

• Model availability may vary by region and API key permissions
• Extended thinking is only available on Opus 4.5+ models
• Newer models generally have improved capabilities
• Model names include version dates for explicit versioning
• Always check Anthropic's documentation for the latest model information
• Enum constants updated with library releases
• String-based names allow forward compatibility with unreleased models
• Context windows are consistent across tiers (200K tokens)
• Max output tokens vary by tier (Opus: 16K, Sonnet: 8K, Haiku: 4K)

Resource Lifecycle:

• Model enum values are constants (no resource management)
• Safe to use across threads and instances
• No memory overhead beyond enum constant itself

Thread Safety:

• Enum constants are inherently thread-safe
• toString() is thread-safe (immutable)
• Safe for concurrent access from multiple threads