Reasoning mode
See a model think before it answers, and know which models can.
Some models can work through a reply before writing it — planning a scene, keeping a mystery consistent, deciding what a character would actually do. Reasoning mode turns that step on and shows it to you in a collapsible pane above the reply.
Which models can reason
| Model | Multiplier | Reasoning |
|---|---|---|
| DeepSeek V4 Flash | 0.3× | Optional |
| MiMo V2.5 | 0.3× | Optional |
| DeepSeek V3.2 | 0.5× | None |
| GLM 5.3 Flash | 0.5× | Always on |
| DeepSeek V4.1 Flash | 0.7× | Optional |
| DeepSeek V4 Pro | 0.7× | Optional |
| DeepSeek R1 | 1.0× | Always on |
| GLM 4.7 | 1.0× | Optional |
| Gemini 3 Flash Preview | 1.2× | None |
| GLM 5 | 1.3× | Optional |
| Llama 3.1 8B | Free | None |
GLM 5.3 Flash and DeepSeek R1 always think before replying and cannot be talked out of it. DeepSeek R1 also has no tool use, so tool-driven plugins won't run on it. GLM 5 and Gemini 3 Flash Preview are advanced models and need a subscription. Full model reference
Turning it on
Side panel → Settings → Replies & Generation → Reasoning Mode, then Enable Reasoning.
The row only exists when the model you're currently on can reason. On an always-on model it's there but locked, reading Reasoning Always On.
The setting is scoped to This device and remembered per model, and it starts off. Switch models and you're back to that model's own setting, not the one you just left.
There is no reasoning toggle in the model picker. The picker has a Reasoning filter chip for narrowing the list and a Reasoning capable badge on the models that qualify. The switch itself lives in chat settings.
Reading the pane
A reasoned reply arrives with a pane headed Reasoning — Reasoning... while it's still coming in — above the reply text. Expand reasoning and Collapse reasoning open and close it.
The pane is collapsed by default. The last message follows your auto-expand setting; older messages keep whatever state you left them in.
Two settings control it, both under Side panel → Settings → Chat Experience, both scoped to This device:
| Setting | Toggle | Default |
|---|---|---|
| Reasoning Pane | Show AI Thinking Process | On |
| Auto-expand Reasoning | Auto-expand for New Messages | Off |
Turn Show AI Thinking Process off if the model's analysis spoils the scene for you — the model still reasons, you just don't see it. With auto-expand off, the waiting indicator reads thinking… rather than typing… while the model works.
What it costs
Reasoning tokens are billed as ordinary completion tokens, at the model's own multiplier. There's no separate reasoning rate and no surcharge.
What does change is volume: a model that thinks first writes more tokens and takes longer to start answering than the same model without it. How much more depends entirely on the model and the scene, so watch your credit balance for the first few replies after switching it on rather than budgeting from a fixed number.
Related
- Choosing a model — which model to reason with in the first place
- Models — capabilities and multipliers for every model
- Side panel & settings — where both reasoning settings live
- Credits — how a reply turns into a charge