#
Auto Mode
One choice in the model selector, the right model for every message.
Experimental
Auto Mode is in active development. The models it uses and how it picks them may change.
Auto Mode picks the model for you. Instead of deciding between a chat, reasoning or multimodal model before each question, you select Auto Mode once and PHOENIQS Chat chooses, message by message, which model handles it. Most messages are answered by a fast, efficient model, and a larger one takes over when needed.
Every model Auto Mode uses runs on the PHOENIQS AI Model Service, so your conversations stay on the same Swiss-hosted infrastructure as when you pick a model yourself.
#
1. Turn on Auto Mode
Open the Model selector and choose Auto Mode under Phoeniqs Experimental, then chat as usual. You can switch back to a specific model at any time, even mid-conversation.
Auto Mode works with what you already use in PHOENIQS Chat: file uploads, images, web search, MCP connectors and agents built with the Agent Builder.
#
2. How it picks a model
For each message, Auto Mode runs a short check-and-improve cycle:
- Draft: an efficient model writes a first answer.
- Review: a second model checks that answer for errors and gaps.
- Hand over, when needed: if the review finds a problem, Auto Mode passes your message to a larger model, and that model writes the answer you receive.
It also adapts to what's in the conversation:
Auto Mode answers in the language you write in, and it knows today's date, so questions like "what's due next week?" work as you'd expect.
#
What you see while it works
While Auto Mode works on your message, a status line shows the current step and the model doing it, for example Checking the answer · GLM 5 Flash · 6s. The steps are:
- Working on it
- Checking the answer
- Checking the change before it runs, before a tool call that changes something
- Handing over to a larger model
When the answer is ready, a summary shows how long it took, how many models were called and how many tool calls were made, for example Worked for 12s · Called 2 models · Made 1 tool call.
#
Models Auto Mode uses
Auto Mode may use the following models from the Active Models catalogue:
inference-deepseek-v32inference-deepseek-v4-flashinference-gemma4-31binference-glm5inference-glm5-flashinference-llama4-maverickinference-llama4-scout-17binference-qwen3-vl-235b
#
3. Quality
Answering with an efficient model only pays off if the answer holds up, so Auto Mode reviews drafts before they reach you. The review looks for real problems rather than style or tone:
- Invented facts: names, numbers, dates or sources the conversation doesn't support.
- Guesswork: gaps in your documents or tool results filled with made-up detail instead of being pointed out.
- Ignored instructions: the format, length or language you asked for.
- Faulty reasoning: arithmetic or logic that doesn't add up, or an answer to a different question.
An honest "your documents don't cover this" counts as a good answer, so Auto Mode doesn't hand over to a larger model just to sound more confident.
The models behind Auto Mode aren't fixed. As new models arrive on the PHOENIQS AI Model Service, we evaluate them and update Auto Mode when they improve results, with no change needed on your side.
Like any review, it won't catch everything. For high-stakes content, switch to a GRC model, which adds guardrails and policies, and check the answer as you would with any model.
#
4. Usage and token credits
Each model call in Auto Mode, including the review, is billed at that model's own token rate and counts towards your token credits, just like any other model. As a rule of thumb, Auto Mode costs less than sending every message to the most capable model, and more than using an efficient chat model on its own. To check what's included in your plan, see Subscriptions.
#
5. When to pick a model yourself
Auto Mode is a good default for everyday work. Choose a specific model instead when:
- You need the same model every time: for example, to compare outputs or follow a workflow your team has validated on one model.
- Speed matters more than a second check: the review adds a few seconds per answer, so a single chat model replies faster for quick back-and-forth.
- You need a GRC model: Auto Mode doesn't use GRC models, so pick one yourself when your work calls for their guardrails and policies.
tip
Not sure where to start? Use Auto Mode for a few days and switch to a specific model only for the tasks where you want full control. The catalogue of recommended models helps you pick one.
#
What's next:
First-time user? → Quick Start
Want Auto Mode to work with your tools? → Connect MCP Tools
Comparing models? → Active Models