Anthropic (Claude)¶

Overview¶

Models: Claude-Opus-4.1, Claude-Opus-4, Claude-Sonnet-4, Claude-3.7-Sonnet, Claude-3.5-Haiku, Claude-3-Opus, Claude-3-Sonnet, Claude-3-Haiku
Features: Chat completions, streaming, system message support

Configuration¶

client, err := omnillm.NewClient(omnillm.ClientConfig{
    Providers: []omnillm.ProviderConfig{
        {Provider: omnillm.ProviderNameAnthropic, APIKey: "your-anthropic-api-key"},
    },
})

Available Models¶

Model	Context Window	Description
`omnillm.ModelClaudeOpus41`	200K	Claude Opus 4.1 (most capable)
`omnillm.ModelClaudeOpus4`	200K	Claude Opus 4
`omnillm.ModelClaudeSonnet4`	200K	Claude Sonnet 4
`omnillm.ModelClaude37Sonnet`	200K	Claude 3.7 Sonnet
`omnillm.ModelClaude35Haiku`	200K	Claude 3.5 Haiku (fast)
`omnillm.ModelClaude3Opus`	200K	Claude 3 Opus
`omnillm.ModelClaude3Sonnet`	200K	Claude 3 Sonnet
`omnillm.ModelClaude3Haiku`	200K	Claude 3 Haiku

TopK Sampling¶

Anthropic supports TopK sampling:

response, err := client.CreateChatCompletion(ctx, &omnillm.ChatCompletionRequest{
    Model: omnillm.ModelClaude3Sonnet,
    Messages: messages,
    TopK: &[]int{40}[0], // Consider only top 40 tokens
})

System Messages¶

System messages are fully supported:

messages := []omnillm.Message{
    {Role: omnillm.RoleSystem, Content: "You are a helpful assistant."},
    {Role: omnillm.RoleUser, Content: "Hello!"},
}

Extended Thinking¶

Anthropic Claude models support extended thinking for complex reasoning tasks. Use the Thinking field for native control:

budget := int64(8192)
response, err := client.CreateChatCompletion(ctx, &omnillm.ChatCompletionRequest{
    Model: omnillm.ModelClaudeSonnet4,
    Thinking: &omnillm.ThinkingConfig{
        Type:         omnillm.ThinkingTypeEnabled,
        BudgetTokens: &budget,
    },
    Messages: []omnillm.Message{
        {Role: omnillm.RoleUser, Content: "Analyze this complex scenario..."},
    },
})

ThinkingType Values¶

Constant	Value	Description
`omnillm.ThinkingTypeEnabled`	`"enabled"`	Enable thinking with explicit budget
`omnillm.ThinkingTypeDisabled`	`"disabled"`	Disable thinking
`omnillm.ThinkingTypeAdaptive`	`"adaptive"`	Let the model decide

Using ReasoningEffort¶

For cross-provider compatibility, you can also use ReasoningEffort, which maps to Anthropic thinking:

ReasoningEffort	Anthropic Behavior
`"none"`	Thinking disabled
`"low"`	Adaptive thinking
`"medium"`	Adaptive thinking
`"high"`	Enabled with budget derived from MaxTokens

effort := omnillm.ReasoningEffortHigh
response, err := client.CreateChatCompletion(ctx, &omnillm.ChatCompletionRequest{
    Model:           omnillm.ModelClaudeSonnet4,
    ReasoningEffort: &effort,
    Messages:        messages,
})

See Reasoning Feature Guide for cross-provider usage.