The four models currently served on the KarlsenAI distributed inference network. Role-based routing across community-run GPUs — try live at ai.karlsen
AI & ML interests
text-generation, image-generation, gguf, quantization, roleplay, uncensored-models, distributed-inference, cryptocurrency, flux, stable-diffusion
Recent Activity
View all activity
Organization Card
KarlsenAI
Distributed AI inference on community-run GPUs. Pay per prompt in KLS or USDT. No signup required to try.
🔗 Try live at ai.karlsencoin.org — free trial (20 prompts/day), no account needed.
What this is
KarlsenAI is a decentralized inference network built on the Karlsen blockchain. Independent GPU operators worldwide run our worker software and get paid directly for each prompt they serve — no middleman, no subscription.
Users pay per request in KLS (Karlsen's native token) or USDT (BEP20). A trial pool sponsored by the project lets anyone try the platform without opening a wallet first.
What's live right now
The network currently serves four models across three roles. Every worker is community-hosted; the routing catalog is public and auditable.
| Role | Model | Notes |
|---|---|---|
| Fast Chat + Vision | Qwen3.5-9B UD-Q4_K_XL |
Native VLM (image upload supported), runs on 8 GB+ GPUs |
| Quality Chat | Qwen3.6-27B UD-Q3_K_XL |
14.8 GB, fits 16 GB cards (RX 6800 class) |
| Quality Chat | Qwen3.8-27B UD-IQ4_XS |
Higher-tier reasoning on 16 GB+ cards |
| Image Generation | Juggernaut XL v9 |
SDXL fine-tune, runs on 8 GB+ NVIDIA/AMD |
All quantizations are the Unsloth Dynamic (UD) GGUFs served via llama.cpp (llama-server), plus SDXL via stable-diffusion.cpp. No proprietary forks.
Why it exists
Most hosted AI services want a credit card, an account, and your prompt history. Most "decentralized AI" projects want a token airdrop and don't ship a working product.
We do neither:
- No signup for trial — 20 free prompts per IP per day, sponsored by the project treasury.
- Direct pay-per-use — every prompt debits your on-chain balance. No monthly plans, no lock-in.
- Prompts and responses are never persisted server-side. Session history lives only in your browser's localStorage.
- Open worker software — anyone with a supported GPU can join the fleet and earn KLS. Source at github.com/karlsencoin/Karlsen-Hybrid.
Run your own worker
If you have an NVIDIA (CUDA) or AMD (ROCm) GPU with 8 GB+ VRAM, you can join the network and earn KLS for each prompt you serve. The single-binary karlsen-hybrid release handles model download, orchestrator handshake, and job execution — Windows and Linux (Ubuntu 22.04+) builds available; HiveOS support in progress.
Download: github.com/karlsencoin/Karlsen-Hybrid/releases (v7.4.0 latest)
Ecosystem
- 🌐 karlsencoin.org — Karlsen blockchain
- 🤖 ai.karlsencoin.org — KarlsenAI live platform
- 🔍 explorer.karlsencoin.org — Karlsen blockchain explorer
- 📊 kgi.karlsencoin.org — Network graph & node map
- 💬 X / Twitter
- 🐙 GitHub
This organization publishes quantizations, sampling presets, and merge experiments used in production on the KarlsenAI network. Everything we publish here is what we actually run.
datasets 0
None public yet