fcfreecompute.orgGet compute

Endpoint

Tweet the site.
Unlock the harness.

Share freecompute.org, send us the tweet URL and your email, and we’ll email your API key when free compute is online. The endpoint is OpenAI-compatible. Two DGX Sparks run DeepSeek V4 Flash around the clock.

API

OpenAI-like

Cluster

2× Spark, 24/7

Limits

30K TPM

Access list

Get access when compute is live

Share freecompute.org publicly, send us the tweet URL and your email. We’ll email your API key when the system is up.

Share freecompute.org on X →

Models

What the pair serves

One checkpoint, tensor-parallel across both Sparks. Same OpenAI client for chat, thinking, and vision.

  • DeepSeek V4 FlashChat

    deepseek-v4-flash

    Default. Tensor-parallel across both Sparks, 24/7.

  • DeepSeek V4 Flash ThinkingReasoning

    deepseek-v4-flash-thinking

    Same weights. Reasoning + output capped at 8K tokens.

  • DeepSeek V4 Flash VisionVision

    deepseek-v4-flash-vision

    Images. Same quota as chat.

Limits

Generous TPM, then we tighten

Starts at 30K TPM, up to 150K TPM with verified followers. We meter input, output, and reasoning on every request, then lower grants slowly as the cluster fills. Follower count never lifts the global cap.

Rate
10 requests / min
In flight
1 request / user
Context
32K max input
Completion
4K max output
FollowersTPM
0–10K30K TPM
10K–100K60K TPM
100K–500K100K TPM
500K+150K TPM

The pair does ~145–350 tok/s non-stop — enough to stay saturated. Extra work queues at 6 cluster-wide jobs. Live context budget is 2M tokens.

OpenAI SDK

Same client, our base URL

https://freecompute.org/v1 · chat.completions · your key when compute is live.

import OpenAI from 'openai'

const client = new OpenAI({
  apiKey: process.env.FREECOMPUTE_API_KEY,
  baseURL: 'https://freecompute.org/v1',
})

const res = await client.chat.completions.create({
  model: 'deepseek-v4-flash',
  messages: [{ role: 'user', content: 'scaffold an app' }],
})