// CHOOSE YOUR ENGINE
Base models
Every model needs a minimum $TOKEN balance to board. Training fees are paid in $TOKEN.
Meta
Llama 3 8B
Lean general-purpose base. Great for chat bots and quick specialists.
- Min hold
- 1,000
- Fee
- 250
- Time
- ~12 min
Mistral AI
Mistral 7B
Punches above its size. Ideal for instruction tuning on a budget.
- Min hold
- 1,000
- Fee
- 220
- Time
- ~10 min
Alibaba
Qwen 2.5 14B
Strong reasoning and multilingual coverage for global audiences.
- Min hold
- 5,000
- Fee
- 600
- Time
- ~25 min
Google
Gemma 2 27B
High quality outputs with balanced compute for production agents.
- Min hold
- 10,000
- Fee
- 1,200
- Time
- ~40 min
Meta
Llama 3 70B
Frontier-class open model for demanding domain experts.
- Min hold
- 50,000
- Fee
- 4,000
- Time
- ~2 hr
DeepSeek
DeepSeek Coder 33B
Purpose-built for code. Train your own repo-aware copilot.
- Min hold
- 25,000
- Fee
- 2,200
- Time
- ~1 hr