recommend

Analysis

Recommend a Model

Find the best Scenario platform/public model for a single generation task. Returns ranked models with real performance data (ELO, latency, cost) and explanations. Args: - capability: e.g. 'txt2img', 'img2img', 'txt2video', 'img2video', 'tts', 'upscale' (optional — inferred from prompt) - prompt: the user's original intent (required) - priority: 'quality' | 'speed' | 'cost' (default 'quality') - max_cost_cu, max_latency_seconds: optional constraints - features: required features list - limit: how many to return (1-10, default 5) Returns: ranked list with per-model explanation, tradeoff, real performance numbers, suggested input params for run_model. When a ranked entry includes `caveats`, surface them to the user before generating with that model — they flag uncertainty (e.g. the model is older than newer alternatives in the same list) that the user should confirm. Don't use when: the request needs multiple generation steps (use plan_generation) or you need to search private/unlisted models (use search).

read-onlyopen-world

Parameters

Name	Type	Required	Description
capability	string	—	Generation capability. Examples: 'txt2img', 'img2img', 'txt2video', 'img2video', 'tts', 'upscale'. If omitted, inferred from the prompt.
prompt	string	—	The user's request. Always pass the user's original intent.
priority	enum(quality \| speed \| cost)	quality	Optimization priority.
max_cost_cu	number	—	Maximum cost per asset in compute units.
max_latency_seconds	number	—	Maximum acceptable generation time in seconds.
features	array	—	Required features (e.g. 'endImage', 'elements').
duration	number	—	Minimum duration in seconds.
limit	number	5
response_format	enum(json \| markdown)	json	Output format: 'json' for structured data, 'markdown' for human-readable text.

Example Request

JSON

{
  "capability": "txt2img",
  "priority": "quality",
  "prompt": "photorealistic product shot for instagram",
  "limit": 3
}

Example Response

JSON

{
  "summary": "For a photorealistic Instagram product shot at 1:1, GPT Image 2 leads on Arena ELO with strong prompt adherence; Gemini 3.0 Pro is the high-fidelity alternative; Grok Imagine Pro adds the best cost-to-quality ratio at this resolution.",
  "ranked": [
    {
      "model_id": "model_openai-gpt-image-2",
      "name": "GPT Image 2",
      "rank": 1,
      "explanation": "ELO 1510 (Arena rank #1, image_edit) and 11.7 CU at 2K make this the top quality choice for photoreal product imagery.",
      "tradeoff": "~65s p50 latency — plan for batch, not real-time.",
      "quality_elo": 1510,
      "speed_summary": "64.7s median",
      "cost_summary": "11.7 CU",
      "suggested_inputs": {
        "aspectRatio": "1:1",
        "numImages": 1
      },
      "key_insights": [
        "Strong prompt adherence on product copy and finishes",
        "Use elementsImage to lock identity across iterations"
      ]
    }
  ],
  "query": {
    "capability": "txt2img",
    "priority": "quality"
  },
  "platform_detected": "instagram_post — 1:1",
  "total_matching": 634
}

Common Use Cases

Pick the best model before calling run_model without scanning the full catalog
Compare quality (Arena ELO), speed (p50 latency), and cost (p50 CU) across candidates side by side
Auto-detect platform context (Instagram, YouTube, TikTok) and receive matching aspect ratio suggestions
Get suggested input values you can pass straight to run_model