Supported Models & Costs

Pocket interacts with OpenAI using the Responses API with strict JSON schema definitions. This ensures every response conforms to Pocket's exact data types without parsing errors or conversational preamble.

Recommended model: gpt-5.4-nano

For almost all users, gpt-5.4-nano is the recommended choice:

• High speed: Returns classification and descriptions in under 1 second.
• Low cost: Costs a fraction of a cent per bookmark. Enriching an active library of 500 bookmarks typically costs less than $0.05 total.
• Strict schema adherence: Consistently respects maximum character constraints (capping titles at 80 characters, descriptions at two sentences, and collections at 18 characters).

Model catalogue

Pocket allows selecting from several models in Settings > Filling in links with AI:

Model	Speed	Cost per 100 Links	Best For
gpt-5.4-nano	Under 1s	~$0.005	Default & Recommended. General bookmark tagging and collection classification.
gpt-4o-mini	Fast (~1s)	~$0.015	Balanced performance across complex foreign language websites.
gpt-4o	Moderate (~2s)	~$0.15	Dense technical and academic articles requiring nuanced summaries.

Structured outputs & privacy

Pocket constructs prompt payloads using OpenAI Structured Outputs:

{
  "model": "gpt-5.4-nano",
  "store": false,
  "response_format": {
    "type": "json_schema",
    "json_schema": {
      "name": "bookmark_enrichment",
      "strict": true,
      "schema": {
        "type": "object",
        "properties": {
          "title": { "type": "string" },
          "description": { "type": "string" },
          "collection": { "type": "string" },
          "tags": { "type": "array", "items": { "type": "string" } }
        },
        "required": ["title", "description", "collection", "tags"],
        "additionalProperties": false
      }
    }
  }
}

• Data minimization: Only the page URL, domain name, existing collection titles, and the first 1,500 characters of clean text are transmitted.
• No storage: Pocket sets "store": false on every API call, preventing OpenAI from saving the request or using your bookmarks for training.