Cheapest vision-capable models suitable for FindRomCover's AI Assist (synchronous chat completions, image input, text/JSON output).

Source: OpenRouter GET /api/v1/models — 446 models total, 275 accept image input (checked 2026-09-20). Prices are USD per million tokens (input / output), ordered by input price. FindRomCover's current OpenRouter default is google/gemma-4-26b-a4b-it.

Estimated cost per 1,000 cover picks (assuming about 6 images ≈ 6k prompt tokens + about 200 completion tokens):

The default google/gemma-4-26b-a4b-it costs about $0.60 per 1,000 picks. See the table above for alternatives.

OpenCode Zen is the OpenCode team's AI gateway. It is OpenAI-compatible and currently offers MiMo-V2.5 Free (mimo-v2.5-free) at no cost for a limited time. MiMo-V2.5 accepts image input, so it works with FindRomCover.

Notes:

• Free for a limited time; Zen states that data from MiMo-V2.5 Free may be used to improve the model while it is free.
• Zen's model metadata may not report image modality, so the picker's Vision-capable only filter can hide it — uncheck the filter or type the model ID directly.
• Zen also serves paid vision models, for example deepseek-v4-flash-vision-exp ($0.14 / $0.28 per M tokens).
• In OpenCode itself the model is referenced as opencode/mimo-v2.5-free; FindRomCover uses the raw model ID mimo-v2.5-free.

• :free variants exist for several models above but are rate-limited and frequently unavailable — use the paid endpoints for batch runs.
• :batch variants are cheaper but asynchronous and do not work with FindRomCover's synchronous client.
• Models without native JSON mode still work: FindRomCover prompts for JSON and parses it from the response text.
• Image token cost is provider-dependent; the estimates above assume about 1,000 tokens per 512px thumbnail.
• Reasoning models may spend output tokens on internal reasoning before producing text. FindRomCover allows up to 4,096 output tokens and reports a clear error if a model still exhausts the budget.

1. Open Settings > AI Settings....
2. Select OpenRouter and paste your API key.
3. Click Test / Load Models.
4. Type a model ID from the table into the filter box and select it.
5. Save, then use AI Pick Best or Batch Fill.

All models listed here are flagged vision-capable by OpenRouter's modality metadata, so they appear when Vision-capable only is checked.

For OpenCode Zen, choose Provider: Custom (OpenAI-compatible), set the Base URL to https://opencode.ai/zen/v1, paste your Zen key, and enter mimo-v2.5-free as the model.

Note: Prices change frequently. Verify current pricing on your provider's website before running large batches.

• Providers & Models
• AI Settings
• Batch Fill