Live thinking
When a model is reasoning through a complex prompt, Anuma streams its thought process to you in real time. You can watch the model work in the activity panel before the final answer arrives.
What live thinking shows
For reasoning-capable models (Claude Opus 4.7, GPT-5 series, o3, o4-mini, DeepSeek, and others), Anuma surfaces the model’s intermediate reasoning steps — the “thinking” the model does before producing a polished response.
This is useful for:
- Trusting the answer — see why the model arrived at a conclusion, not just the final claim
- Catching mistakes early — spot a wrong assumption mid-stream and stop or redirect before wasting credits
- Learning from the model — understand how it breaks down a problem step by step
- Debugging code or math — watch the model trace through logic before committing to an answer
Steps
Step 1: Send a prompt that benefits from reasoning
Live thinking activates automatically for reasoning-capable models on prompts that engage their extended-thinking mode. Examples: multi-step math, complex code debugging, Deep Research queries, or anything you’d expect to take more than a few seconds to answer.
Step 2: Open the activity panel
The activity panel on the right side of the chat shows the model’s live reasoning as it streams. You’ll see thinking tokens appear before the final response starts rendering.
Step 3: Watch or wait
You can read along while the model thinks, or close the panel and wait for the final answer. Live thinking doesn’t change how long the response takes — it just makes the work visible.
Good to know
- Live thinking is available on reasoning-capable models only. For non-reasoning models (most “Fast” variants, Gemini Flash, etc.), you’ll see the final response stream directly without an intermediate thinking step.
- Thinking tokens are part of the model’s compute and are reflected in the credit cost. Reasoning-heavy responses cost more than chat-style responses.
- The activity panel also shows tool calls, file uploads, and search activity while the model works — not just thinking.
- You can hide the activity panel anytime; live thinking continues to stream in the background.
Related articles