Hugging Face is out. Who do you actually sign up with today?

Hugging Face exit sheet — who to sign up with (2026-08-27)

Someone asked on HN today: "Hugging Face is out. Where can I sign up to move away from HF?"
Thread: https://news.ycombinator.com/item?id=49465640

This is a same-day public scrape of signup + pricing pages. Prices move. Click the pricing URL before you commit.

Fastest swap (OpenAI-compatible API)

1. Together AI — https://api.together.ai
   Pricing: https://www.together.ai/pricing
   Llama 3.3 70B: $1.04 in / $1.04 out per 1M tokens (their page today)
   gpt-oss-120B: $0.15 / $0.60 per 1M
   Dedicated H100: $5.49/GPU-hr on-demand. Cluster H100: $3.99/hr
   Start: serverless, then dedicated if you need a box.

2. Fireworks AI — https://fireworks.ai
   Pricing: https://fireworks.ai/pricing
   $1 free credit on signup. Serverless is per-token (model table is in their docs).
   On-demand H100: $7/hr through Aug 31, $8/hr from Sep 1 (their page today)

3. Groq — https://console.groq.com
   Marketing site: https://groq.com
   Fastest for the models they actually list. Do not assume every HF weight is there.

You want a GPU, not a token API

1. Runpod — https://console.runpod.io/deploy Pricing: https://www.runpod.io/pricing H100 PCIe: $1.99/hr community / $2.89/hr secure (page today) H100 SXM: $2.69 / $3.29 RTX 4090: $0.34 / $0.74 You run the container. Cheapest raw box. You own the serving stack.

Also worth a tab

1. OpenRouter — https://openrouter.ai — one key, many hosts
2. Replicate — https://replicate.com — pay-per-second community models
3. Modal — https://modal.com — Python, no YAML
4. Baseten — https://www.baseten.co — dedicated custom/fine-tune
5. Nebius Token Factory — https://nebius.com — dedicated endpoints
6. DeepInfra — https://deepinfra.com — cheap OpenAI-compatible

10-minute move

# old
# HUGGINGFACE_URL + HF token

# new (Together example)
from openai import OpenAI
client = OpenAI(base_url="https://api.together.xyz/v1", api_key="TOGETHER_KEY")
client.chat.completions.create(
    model="meta-llama/Llama-3.3-70B-Instruct-Turbo",
    messages=[{"role":"user","content":"ping"}],
)

Swap the base_url + model id. Keep the rest of your code.

Pick one

• Chat API this afternoon → Together or Fireworks
• Lowest latency on a short model list → Groq
• Fine-tune + serve the same stack → Fireworks or Together
• Your own weights on a box → Runpod
• Do not want to pick a host → OpenRouter

Sources (fetched 2026-08-27):

• https://www.together.ai/pricing
• https://fireworks.ai/pricing
• https://www.runpod.io/pricing
• https://groq.com/pricing