Choosing a Backend by Which Modality Your Input Actually Is: Text-Only, Image-Bearing or PDF-Bearing Prompts on One Key

Why modality matters first

When you have one API key that unlocks several backends, the fastest way to choose is to start with what your prompt actually contains. Text, images, and PDFs each impose different requirements on the model. Some backends handle all three; others are optimized for text alone. By filtering on modality first, you avoid trial-and-error and keep costs predictable.

Know your input modality

Before comparing backends, classify your prompt:

• Text-only: The prompt is plain text, possibly with structured formatting (Markdown, JSON, code). No files attached.
• Image-bearing: The prompt includes one or more images (screenshots, photos, diagrams) that the model must interpret.
• PDF-bearing: The prompt includes a PDF document, which may contain text, tables, and images. PDFs often require the model to extract and reason over multiple pages.

Each modality narrows the set of suitable backends. The next sections walk through the filtering logic.

Filtering for text-only prompts

Text-only prompts are the most widely supported. Almost every backend accepts them, so your choice can focus on other factors:

• Task type: Some backends excel at code generation, others at creative writing or long-context reasoning.
• Context length: If your text is very long, check which backends support large context windows.
• Cost sensitivity: All else equal, you pay the same multiplier (official price ×1.3) regardless of backend, but the base official price varies by model. Compare the underlying official prices to estimate your USDC cost.

Because text-only support is near-universal, you can shortlist based on quality and price rather than availability.

Filtering for image-bearing prompts

When your prompt includes images, you need a backend that accepts image inputs (often called vision or multimodal capability). Not all backends offer this.

• Check modality support: Review the documentation for each backend to confirm it accepts images. Some may support images only in certain model variants.
• Image format and size: Confirm that the backend accepts your image format (e.g., JPEG, PNG) and that the image dimensions are within limits. Large images may be downscaled or rejected.
• Cost implications: Image inputs are typically tokenized differently and may increase the effective prompt size. Your USDC cost will reflect the official price for image tokens multiplied by 1.3. If you are a contributor, you receive credits at official price ×1.1 (or ×1.2 for premium models).

If image support is critical, eliminate backends that do not list it before comparing other features.

Filtering for PDF-bearing prompts

PDFs are more complex. Some backends accept PDFs directly, while others require you to extract text or convert pages to images first.

• Direct PDF support: Check whether the backend accepts PDF files as input. If it does, confirm any page limits or file size restrictions.
• Indirect approach: If a backend does not accept PDFs, you can often convert PDF pages to images and send them as image-bearing prompts—provided the backend supports images. Alternatively, extract text locally and send a text-only prompt, though this may lose layout information.
• Cost considerations: PDFs can be token-intensive. A multi-page PDF may consume many tokens, so estimate the official price and apply the ×1.3 multiplier for your USDC cost. Contributors earn credits at ×1.1 (or ×1.2 for premium) of the official price.

PDF support is less common than image support, so this filter can quickly narrow your options.

Putting it together: a decision walkthrough

Given a single prompt, follow these steps:

1. Classify the modality: Is it text-only, image-bearing, or PDF-bearing?
2. Eliminate backends that lack support: For images, remove text-only backends. For PDFs, remove those without direct PDF support (unless you plan to convert).
3. Compare remaining backends on cost and capability: Use official prices to estimate your USDC cost (×1.3). Consider context length, model strengths, and any premium tiers (which may affect contributor credits).
4. Send your prompt with one key: Your API key works across all supported backends, so you can switch without new credentials.

This modality-first approach keeps your selection grounded in what the backend can actually do, not just marketing claims.

Understanding the USDC cost model

All backends share the same pricing structure:

• Users pay the official price multiplied by 1.3, settled in USDC on Base. No KYC is required.
• Contributors (those who add value, e.g., by providing compute or models) receive credits at the official price multiplied by 1.1, or ×1.2 for premium models, paid in USDC.

This means your cost is predictable: take the official price for the model and modality, apply the multiplier, and you have your USDC cost. There are no hidden fees or variable rates beyond the official price.

Practical tips

• Start with modality, then optimize: Don't compare backends on price until you've filtered by modality.
• Test with small prompts: Before sending a large PDF or high-resolution image, try a minimal example to confirm the backend handles your format.
• Monitor token usage: Images and PDFs can consume more tokens than plain text; keep an eye on the official price to estimate your USDC spend.
• Leverage contributor credits: If you contribute, your effective cost is lower because you earn credits at ×1.1 or ×1.2 of the official price.

By following this walkthrough, you can confidently pick the right backend for each prompt without wasting time or USDC.