AI Horde
Free, volunteer-run, open-source generation network. Community workers serve Heretic and uncensored models, and prompts may be visible to whoever runs the worker.
BILLINGFree (crowdsourced volunteer workers). Priority via kudos (earned by contributing). 36 text models / 44 workers online at snapshot; list changes hourly. Other non-uncensored models online: google/gemma-4-31b (count 4, 262144 ctx), Llama-3.2-3B/1B-Instruct, gemma-3-1b/4b-it, Qwen3.5-0.8B, Qwen3-0.6B, Qwen2.5-Coder-3B, gemma-4-E2B variants, Gemma-4-26B-A4B-StyleTune-V2.
| Model | Context |
|---|---|
| 66K | |
| 16K | |
| 16K | |
koboldcpp/L3-8B-Stheno-v3.2Llama-3-8B · 8B | 8K |
| 16K | |
koboldcpp/Rocinante-X-12BMistral-Nemo-12B · 12B | 33K |
koboldcpp/gemma-4-31B-it-hereticGemma-4-31B-it · 31B · Apache-2.0 | 33K |
koboldcpp/Gemma-4-E4B-Uncensored-HauhauCS-Aggressive-Q4_K_MGemma-4-E4B · 4B · Apache-2.0 | 12K |
koboldcpp/G4-MeroMero-26B-A4B-it-uncensored-hereticGemma-4-26B-A4B-it · 26B · Apache-2.0 | 66K |
koboldcpp/ChatML-thinking-llmfan46-Qwen3.6-35B-A3B-uncensored-hereticQwen3.6-35B-A3B · 35B | 45K |
koboldcpp/Gemma-4-E4B-it-Ultra-Uncensored-HereticGemma-4-E4B-it · 4B · Apache-2.0 | 33K |
koboldcpp/toandev/Gemma4-12B-UncensoredGemma-4-12B · 12B · Apache-2.0 | 10K |
koboldcpp/TheDrummer/Cydonia-24B-v4.3Mistral-Small-24B · 24B · Apache-2.0 | 20K |
koboldcpp/Cydonia-v1.3-Magnum-v4-22BMistral-Small-22B · 22B · Apache-2.0 | 8K |
koboldcpp/gemma-3-4b-it-hereticGemma-3-4B-it · 4B | 4K |
koboldcpp/Mistral-Nemo-2407-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETICMistral-Nemo-12B-2407 · 12B | 1K |
| 49K |
- Method
- Mixed
- Access
- API
- Logging
- Retained
- Trains on prompts
- Unverified
- Guarantee
- Policy only
- Jurisdiction
- Not disclosed
- Signup
- No personal info
- Payments
- Unverified
- Data checked
- Oct 11, 2026
PRIVACYThe central server holds prompts only transiently, but volunteer workers can technically see them. The project's own FAQ advises treating every request like a post in a public forum. Policy ↗
35 text models were live on 2026-10-11, and availability depends on which volunteers are online. Use is free; "kudos" only set queue priority. Chat is through third-party front-ends such as KoboldAI Lite.