Skip to main content
The HASP AI API provides access to three model tiers. Each is a direct Anthropic model, routed through the HASP compliance substrate.

​
Available models

Model identifierAnthropic modelCharacteristics
claude-haiku-4-5claude-haiku-4-5-20251001Fastest, lowest cost. Best for high-volume tasks, simple classification, short generations.
claude-sonnet-4-6claude-sonnet-4-6Balanced speed and capability. Default recommendation for most production workloads.
claude-opus-4-6claude-opus-4-5Highest capability. Requires explicit opt-in. Best for complex reasoning, long-context synthesis, high-stakes decisions.
Pass the model identifier in your request: 
{
  "model": "claude-sonnet-4-6",
  "messages": [{ "role": "user", "content": "..." }]
}

​
Opus opt-in

Access to claude-opus-4-6 requires explicit enablement on your organization. Requests to Opus without enablement fail with: 
{
  "error": {
    "code": "OPUS_NOT_ENABLED",
    "message": "Opus access requires explicit enablement. Enable it in Settings → AI Workspace → Models."
  }
}
Enable Opus under Settings → AI Workspace → Models in the HASP dashboard. The enablement is per-organization; once enabled, any API key with ai:chat scope can use Opus.

​
Cost comparison

Credit consumption scales with model capability. As a rough guide:
  • Haiku is ~5x cheaper than Sonnet per token.
  • Sonnet is ~5x cheaper than Opus per token.
See Budget Controls for exact credit rates. For most regulated-industry workloads — clinical documentation assistance, compliance summarization, patient communication drafts — Sonnet delivers the best capability/cost tradeoff. Use Haiku for preprocessing, classification, and routing. Reserve Opus for final reasoning steps where accuracy is the primary constraint.

​
Deprecation policy

Model versions are deprecated with 6 months advance notice and a 12-month overlap period during which both the old and new versions are available. The model identifier you pass in the request always resolves to a specific model version — HASP does not silently upgrade you to a newer version without your opt-in. When a model version is retired, requests using the retired identifier return: 
{
  "error": {
    "code": "MODEL_RETIRED",
    "message": "Model claude-X-Y-retired was retired on 2026-MM-DD. Migrate to claude-X-latest."
  }
}
Subscribe to the Changelog to receive advance notice of upcoming retirements.