← back to homeTwo models.
// models
Two models.
One uncensored agent.
TartarusAI ships two custom-tuned coding models on dedicated GPU infrastructure. Both inherit the same uncensored posture and the same runtime safety guards. You pick the one that fits the task — Coder for generation, Thinking for reasoning.
Coder
read more →MoE 30B total · 3B active per token
Default coding model — code generation, refactor, multi-file edits
- ●Lower TTFT and higher throughput than the Thinking variant
- ●Tuned on offensive-security and adversarial-code workflows
- ●Handles 256K context on Pro+ tier
Thinking
read more →MoE 30B total · 3B active · reasoning chain enabled
Reasoning variant — planning, debug analysis, refactor decisions, architectural review
- ●Visible reasoning trace for audit and verification
- ●Stronger on multi-step planning where Coder can over-commit
- ●Same context window; pick by task, not by capability