← back to models
// model · thinking
Thinking model.
Reasoning-mode variant of the same MoE 30B/3B architecture. Visible reasoning trace (think blocks) makes the agent\\'s plan auditable before it commits to code. Switch to Thinking when the task is "what should we do" rather than "implement this."
- Architecture
- MoE · 30B total · 3B active
- Context window
- 128K (Pro) · 256K (Pro+)
- Reasoning trace
- Yes · visible <think> blocks
- Strengths
- Planning, debug, refactor decisions
- TTFT
- Higher than Coder · trade for depth
- Best for
- Architectural review, exploit chains
Thinking is the right pick for ~20% of workflows where you want to see the agent\\'s plan before code commits. For most generation work, Coder is faster.