Skip to main content
Unlayer Docs
Unlisted page
This page is unlisted. Search engines will not index it, and only users having a direct link can access it.

Picking a Model

features.ai.model controls the preferred model for the AI Assistant's text and chat — generating and editing copy, layouts, and design. It is independent of features.ai.image.model, which only affects AI image generation and editing.

Recommended: let Unlayer pick

Leave features.ai.model unset and Unlayer picks the best model per request. Set it only when you need to pin a specific provider or model.

Pin a model​

features.ai.model accepts:

  • a "<provider>/<modelId>" pair (e.g. "openai/gpt-5.6-luna", "anthropic/claude-opus-5") — pins both provider and model, unambiguous;
  • a bare model id (e.g. "gpt-5.6-luna");
  • a provider name (e.g. "openai", "anthropic") — starts on that vendor and lets Unlayer pick that provider's default model.

Pinned models are treated as the first choice for the request. If the selected model or provider is rate-limited, overloaded, unavailable, or returns another transient provider error before the user sees output, Unlayer may retry the turn against a backup model so the assistant can stay available during provider incidents.

unlayer.init({
  features: {
    ai: {
      enabled: true,
      assistant: true,

      // Provider only — Unlayer picks that provider's default model:
      model: 'anthropic',

      // Or pin an exact model:
      // model: 'openai/gpt-5.6-luna',
    },
  },
});

Supported providers: openai, anthropic.

Configure fallback models​

features.ai.fallbackModels controls the transient-outage fallback tail:

  • omit it to use Unlayer's default fallback picks only when features.ai.model is unset;
  • set true to use Unlayer's default fallback picks even when features.ai.model pins a provider or model;
  • set false to disable the default outage fallback tail;
  • pass an ordered model array to replace the default fallback picks.

Fallbacks are only used for provider availability failures such as network errors, overloads, rate limits, and 5xx responses. They are not used for invalid request bodies, invalid API keys, authorization errors, credit exhaustion, or errors that require a change from you or Unlayer.

unlayer.init({
  features: {
    ai: {
      enabled: true,
      assistant: true,
      model: 'anthropic/claude-opus-5',

      // Try these only if the preferred model has a transient outage:
      fallbackModels: ['anthropic/claude-opus-4-8', 'openai/gpt-5.5'],

      // Or use Unlayer's default outage tail after this pinned model:
      // fallbackModels: true,

      // Or disable default outage fallbacks:
      // fallbackModels: false,
    },
  },
});

Quality vs. cost​

If you do pin a model, here's how the tiers generally compare:

TierExamplesRelative costBest for
Frontierclaude-opus-5, gpt-5.6-solHigherDesign-heavy work where maximum visual taste and copy quality justify the premium.
Balancedclaude-sonnet-5, gpt-5.6-terraStandardEveryday design and content work with a balance of capability, latency, and cost.
Efficientclaude-haiku-4-5, gpt-5.6-lunaLowerHigh-volume agentic work where minimizing cost matters more than maximum quality.

With automatic routing, straightforward design turns start on Terra with medium reasoning effort. More demanding requests can escalate to Opus, while specialized flows such as text editing and imports may use their own defaults. Templates that declare custom tools start on Anthropic because custom tool schemas are not yet guaranteed to work across every provider. An explicit model selection still takes precedence.