Models

List available models from AI provider APIs with automatic caching.

First of Its Kind

Neither Prism nor any other PHP AI SDK provides a way to list available models from providers. Atlas is the first to ship this capability.

Supported Providers

Provider	Supported	Format
OpenAI	Yes	OpenAI-compatible
Anthropic	Yes	Anthropic
Gemini	Yes	Gemini
Ollama	Yes	OpenAI-compatible with native fallback
DeepSeek	Yes	OpenAI-compatible
Mistral	Yes	OpenAI-compatible
Groq	Yes	OpenAI-compatible
XAI	Yes	OpenAI-compatible
OpenRouter	Yes	OpenAI-compatible
ElevenLabs	Yes	ElevenLabs
Perplexity	No	No models endpoint
VoyageAI	No	No models endpoint
Z	No	No models endpoint

Basic Usage

use Atlasphp\Atlas\Atlas;

// List models from a specific provider
$models = Atlas::models('openai')->all();

// Returns a simple sorted list of model identifiers
foreach ($models as $model) {
    echo $model; // e.g., "gpt-4o"
}

Using Prism Provider Enum

use Prism\Prism\Enums\Provider;

$models = Atlas::models(Provider::Anthropic)->all();

Check Provider Support

if (Atlas::models('elevenlabs')->has()) {
    // Won't reach here — ElevenLabs has no models endpoint
}

Caching

Model lists are cached by default since they change infrequently. Configure in config/atlas.php:

'models' => [
    'cache' => [
        'enabled' => true,    // Enable/disable caching
        'store' => null,      // null = default cache store
        'ttl' => 3600,        // Cache lifetime in seconds
    ],
],

Force Refresh

// Bypass cache and fetch fresh from the API
$models = Atlas::models('openai')->refresh();

Clear Cache

// Clear cached models for a specific provider
Atlas::models('openai')->clear();

Return Format

All methods return list<string> — a sorted list of model identifier strings (e.g., ['gpt-3.5-turbo', 'gpt-4o']). These are the identifiers you pass to ->using() when making API calls.

OpenAI-Compatible Providers

Most providers (Groq, DeepSeek, Mistral, XAI, OpenRouter, Perplexity, VoyageAI) use the OpenAI-compatible /v1/models endpoint format. If you're using a custom OpenAI-compatible provider, it will work automatically as long as it follows the same response format.

API Reference

// All methods require a provider
Atlas::models(Provider|string $provider)

// Get models (cached or fresh)
->all(): ?array

// Check if provider supports model listing
->has(): bool

// Force refresh from API
->refresh(): ?array

// Clear cached models
->clear(): void

Pipeline Hooks

Model listing does not currently participate in the pipeline system. This may be added in a future version if observability hooks around model fetching prove useful.

Next Steps

• Chat — Use listed models for chat requests
• Embeddings — Use listed models for embeddings
• Pipelines — Add observability hooks