Open-source tool pxpipe hides text in PNGs to cut Claude Code and Fable 5 token costs up to 70%

What changed

Pxpipe is an open-source tool designed to cut costs for users of Anthropic’s Claude Code and Fable 5 models by encoding long text prompts into PNG images. The trick exploits how Anthropic charges for image inputs based on pixel size rather than the amount of text inside. Instead of sending verbose prompts as costly tokens, pxpipe compresses them into compact PNG files, which get counted differently for billing purposes.

Why builders should care

Text prompt length can balloon costs quickly when working with Claude Code and similar models. For founders and developers relying on these LLM APIs, pxpipe’s approach slashes expenses by 59 to 70 percent according to developer Steven Chong. That hits the bottom line hard and could make more complex prompt engineering financially feasible. It also invites rethinking how prompt input formats influence pricing strategies for text-to-LLM workflows.

The practical takeaway

Users willing to trade off some accuracy and speed can save substantially by swapping raw text inputs for PNG-encoded prompts. This means bigger, richer context windows without a linear cost increase in tokens paid. Builders running frequent or heavy prompt calls need to evaluate pxpipe integration if they want to optimize spend under Anthropic’s pricing model. However, the compromises on model response fidelity and latency require careful testing—cost savings come with operational trade-offs.

What to watch next

Watch for wider uptake of encoding tricks like pxpipe that lean on pricing quirks in multi-modal and text-to-LLM APIs. Anthropic and other providers may adjust pricing to close these loopholes, potentially shifting costs back onto users. Developers should monitor updates from pxpipe and evaluate alternative cost-cutting approaches. Also track any improvements in decoding accuracy and speed by pxpipe that could make the tool more broadly practical.

AI Quick Briefs Editorial Desk