DEV Community

Diego Siguelnitzky
Diego Siguelnitzky

Posted on

Why we built 50X: Low-cost inference, starting with DeepSeek, powered by distributed data centers.

Most people using DeepSeek V4.1 Flash today are either:

  1. Hitting the official API (data processed and stored in China, no zero-retention option), or
  2. Paying OpenRouter ~5× more for the same open weights.

Neither is ideal if you’re running agents that stay up for hours, burn a lot of tokens, and handle real code or customer data.

We started 50X because we wanted a third option:

  • The open weights of DeepSeek V4.1 Flash
  • Hosted on our own GPUs
  • Zero retention (we don’t train on your data and we don’t keep prompts)
  • $0.03 / $0.20 per million tokens (input / output)
  • OpenAI-compatible endpoint

That’s it. No new model, no fine-tune, no claim that we’re faster than everyone. Just the model people already want, served in a way that makes long-running agent work actually affordable and less risky on the data side.

Who this is for

  • People building agents that run unattended for hours
  • Teams that need cheap coding/inference but don’t want their prompts sitting in China
  • Anyone already using OpenRouter or the official DeepSeek API and looking at the bill

How to try it

OpenAI-compatible base URL:
https://api.ai.50x.energy/v1

$100 in free credits for the first 100 accounts (no card).

Docs and quickstart: https://50x.energy

If you’re already hitting a limit on free tiers or watching your OpenRouter spend climb, this is the drop-in alternative we built. Happy to answer questions in the comments about context length, quantization, or how we handle long sessions.

Top comments (0)