Claude Haiku 4.5

model: 'claude-haiku-4-5'

Available through anthropic, vertex, and openrouter. The fastest and most cost-effective Claude model maat supports — a good default for high-volume enrichers.

Configuration

// Anthropic
{
  provider: 'anthropic',
  model: 'claude-haiku-4-5',
  extra: { apiKey: process.env.ANTHROPIC_API_KEY },
}

// Vertex AI
{
  provider: 'vertex',
  model: 'claude-haiku-4-5',
  extra: { project: 'my-gcp-project', location: 'us-central1' },
}

See Anthropic, Google Vertex AI, or OpenRouter for authentication.

Specifications

Max input tokens	200,000
Max output tokens	64,000
Temperature	Not sent — sampling parameters are unsupported on this model

Note the smaller 200,000-token input window — the rest of maat's models accept 1,000,000.

Cost

	Cost
Input tokens	$1.00 / million
Output tokens	$5.00 / million

Costs are estimates. Actual billing depends on the provider and region.

JSON output

JSON enrichers use schema-enforced structured outputs — output_config.format on the native Anthropic and Vertex paths, response_format: json_schema via OpenRouter.

Timeout

timeoutMs applies per individual LLM call within a batch. If unset, no explicit timeout is applied beyond the provider's default.

Related

• Anthropic
• LLM models overview