AI Models Settings
The AI Models Settings page lets you choose the workspace model and configure how employees carry out work and delegate.
TL;DR
Go to Settings > Technical > AI Models to browse models by tier and set workspace-wide work behavior. The model you pick here is what every employee uses. The tier is a rough capability class, and credit cost per message varies by the specific model you choose.
Where to Find It
- Click the gear icon in the top-right of your workspace
- Go to Technical > AI Models
Model Boost
When Model Boost is available, new workspaces get two boosts by selecting a workflow or playbook from the Work menu. Ordinary chat and typed requests use the usual model.
Each boost covers the employee's initial reply and its primary model calls while preparing that reply. Later workflow steps and delegated work use their usual models. A boost does not cover an entire workflow, and its model calls use your credits.
Existing workspaces keep their original reply-based allowance and usage. The counter on the AI Models page shows your workspace's allowance. Turning Model Boost off pauses the remaining allowance; turning it back on does not reset it. A reply attempt uses one boost even if it fails before a reply is delivered. Retrying that same saved message does not use another boost.
Model Tiers
Every model belongs to one of four tiers. The tier is a capability class, not a strict price ladder: a Fast model and an Advanced model can land close in credit cost, so check the exact rate on the model you pick.
| Tier | Capability | Typical Use |
|---|---|---|
| Fast | Good for simple, well-defined work | Quick lookups, formatting, simple drafts, routine Q&A |
| Standard | Handles most everyday tasks well | Conversations, analysis, writing, planning |
| Advanced | Strong reasoning and nuance | Complex analysis, long documents, multi-step reasoning |
| Frontier | The top-of-line brains | The hardest work, where quality matters most |
Available Models
The models below reflect the current selectable catalog. Your available choices depend on workspace access and any model-cost ceiling. Check the AI Models page for current rates and locked choices.
Fast Tier
| Model | Provider |
|---|---|
| Gemini 3.1 Flash Lite | |
| Gemini 3.5 Flash Lite |
Standard Tier
| Model | Provider |
|---|---|
| DeepSeek V4 Flash | DigitalOcean |
| Gemini 3.5 Flash | |
| GLM 5.3 Flash | Relace |
| GLM 5.3 FlashX | Z.AI |
| GPT-5.4 Mini | OpenAI |
| GPT-5.6 Luna | OpenAI |
| GPT-6 Luna | OpenAI |
| Grok 4.3 | xAI |
| Haiku 4.5 | Anthropic |
| Kimi K2.5 | DeepInfra |
| Kimi K2.6 | DeepInfra |
| Llama 4 Maverick | Parasail |
| MiniMax M2.7 | DeepInfra |
| o4 Mini | OpenAI |
| Qwen 3.8 Flash | Alibaba |
| Qwen3 Coder | DeepInfra |
Advanced Tier
| Model | Provider |
|---|---|
| DeepSeek R1 | SiliconFlow |
| DeepSeek V4 Pro | DeepSeek |
| DeepSeek V4 Pro 0813 | Baidu |
| DeepSeek V4.1 Flash | DeepInfra |
| Gemini 3.1 Pro | |
| Gemini 3.6 Flash | |
| Gemini 3.7 Flash | |
| Gemini 3.8 Flash | |
| GLM 5.2 | DeepInfra |
| GLM 5.3 | DeepInfra |
| GLM 5.3 Prime | Alibaba |
| GPT-5.2 | OpenAI |
| GPT-5.4 | OpenAI |
| GPT-5.6 Terra | OpenAI |
| GPT-6 Sol | OpenAI |
| GPT-6.1 Sol | OpenAI |
| GPT-6.1 Terra | OpenAI |
| Grok 4.5 | xAI |
| Grok 4.6 | xAI |
| Grok 4.7 | xAI |
| MiniMax M3 | MiniMax |
| MiniMax M3.1 Flash | MiniMax |
| Mistral Large | Mistral |
| Mistral Medium 3.5 | Mistral |
| Qwen 3.7 Plus | Alibaba |
| Qwen 3.8 2.4T | Novita |
| Qwen 3.8 Max | Alibaba |
| Qwen 3.8 Max Prime | Alibaba |
| Qwen3 235B Thinking | Alibaba |
| Sonnet 4.6 | Anthropic |
| Sonnet 5 | Anthropic |
| Sonnet 5.5 | Anthropic |
Frontier Tier
| Model | Provider |
|---|---|
| Fable 5 | Anthropic |
| Fable 5.1 | Anthropic |
| GPT-5.2 Pro | OpenAI |
| GPT-5.4 Pro | OpenAI |
| GPT-5.5 | OpenAI |
| GPT-5.5 Pro | OpenAI |
| GPT-5.6 Sol | OpenAI |
| GPT-6 Astra | OpenAI |
| Kimi K3 | Moonshot AI |
| Kimi K3.1 Pro | Moonshot AI |
| Opus 4.5 | Anthropic |
| Opus 4.6 | Anthropic |
| Opus 4.7 | Anthropic |
| Opus 4.8 | Anthropic |
| Opus 5 | Anthropic |
| Opus 5.5 | Anthropic |
Retired Models
Existing selections move to their replacement; historical usage remains visible. These are Sistava retirement dates, which may differ from provider shutdown dates.
| Retired model | Sistava retirement date | Replacement |
|---|---|---|
| GPT-4.1 Nano | November 2, 2026 | GPT-6 Luna |
| Claude Opus 4.1 | November 9, 2026 | Claude Opus 5.5 |
| Claude Sonnet 4.5 | November 9, 2026 | Claude Sonnet 5.5 |
| GPT-5 Mini, GPT-5.4 Nano | November 9, 2026 | GPT-6 Luna |
| GPT-4o Mini, GPT-4.1 Mini | November 9, 2026 | GPT-6 Luna |
| GPT-4o, GPT-5.1 | November 9, 2026 | GPT-6.1 Terra |
| GPT-5 Nano | November 9, 2026 | GLM 5.3 Flash |
What You See on the Page
The AI Models Settings page displays all available models grouped by tier in the Employee Brain card. For each model you can see:
- Model name and provider
- A 1 to 5 star rating for reasoning capability
- A brain icon, if the model streams a visible thinking process
- Recommended or Expensive badges, where they apply
- The exact credit cost per million tokens
You can also sort the list by cheapest, by provider, or by reasoning strength instead of by tier, and filter to reasoning-only models.
Setting the Model
The model you pick here is what every employee in your workspace uses. There is currently no per-employee override in the interface: model, thinking depth, and per-turn action limit are controlled workspace-wide only, to keep one place in sync instead of two.
- Go to Settings > Technical > AI Models
- In the Employee Brain card, pick the model you want
- The change saves automatically and takes effect on each employee's next turn
When your workspace has a model-cost ceiling, choices above it show a lock and upgrade prompt. Models within the permitted limit remain selectable. Workspaces with full selection access can choose from the available catalog; use the picker as the current source for your workspace.
Work Controls
How Employees Work and Delegation & Teamwork are workspace-wide settings. They include Actions per Turn, Loop Protection, Parallel Delegations, Delegation Reach, and Teammate Work Time.
Every numeric control shows an exact value and an allowed-range indicator. That indicator shows the setting you selected between the policy minimum and maximum. It is not a quota meter and does not claim to show capacity remaining. The plan-capacity meters for credits, employee slots, and email sends remain on Rate Limits.
Conversation Continuity
One card groups all recent-conversation controls. Changes save automatically for the workspace.
| Conversation source | Setting | Default | Allowed range |
|---|---|---|---|
| You and the current employee | Recent messages from this chat | 30 | 0–60 |
| You and other employees | Other employees to include | 7 | 0–10 |
| You and other employees | Messages per employee | 3 | 0–10 |
| Employee-to-employee conversations | Total messages | 30 | 0–60 |
| Employee-to-employee conversations | Messages per teammate | 15 | 0–30 |
Other employees are selected by when you last spoke to them. Their excerpts include your messages and their replies, with dates and speaker names, across your eligible private threads. The current employee is excluded from that list. Long messages are shortened, so these counts are ceilings, not complete transcripts. Employee-to-employee exchanges also have a total text limit; both total-message and per-teammate limits apply.
Your private chats stay tied to you. Another human speaking to the same employee does not receive these excerpts from your chats. Membership and source access are checked. Archived and older shared threads are excluded from the cross-employee view. These settings do not grant anyone additional access.
Choose 0 to omit the corresponding excerpts. Either zero employees or zero messages turns off the other-employee view. Either zero total messages or zero per teammate turns off teammate history. This does not delete conversations, clear saved thread state, disable saved notes, or switch off Advanced Memory. See Memory for those separate mechanisms.
Advanced: Choosing Each Setting
| Control | When to adjust it | Tradeoff |
|---|---|---|
| Model | Match the workspace model to your work | Capability, speed and credits vary. Available choices follow workspace access and cost limits. |
| Thinking Depth | Deeper for difficult reasoning; lighter for routine work | Deeper can take longer and use more credits. Disabled when the model does not support this control. |
| Creativity | Lower for steady wording; higher for varied drafts | More variation does not guarantee accuracy. Range 0–2, usual default 0.7. |
| Credits Saver | Use a cheaper model for simple conversation | Real work keeps the chosen model. Review tracked savings on its linked page. |
| Model Boost | Pause or resume remaining initial-reply boosts | Does not reset used boosts. Boost model calls still use credits. |
| Actions per Turn | Increase when legitimate work reaches the turn limit | Allows longer, more expensive runs. Override access depends on your plan. |
| Loop Protection | Stricter for repeated tool calls; relax when repetition is needed | Stricter protection can stop legitimate repeated work earlier. |
| Parallel Delegations | Let a leader handle more teammate assignments at once | More concurrent work can spend credits faster. |
| Delegation Reach | Choose solo, same-team or cross-team work | Broader coordination does not bypass source access. |
| Teammate Work Time | Allow longer delegated tasks | Longer timeouts allow work and costs to continue longer. |
| Messages from this chat | Increase if the recent excerpt misses important context | More model input; separate from saved thread state. |
| Other employees to include | Increase when work often moves between employees | More context and input cost; reduce for a narrower view. |
| Messages per other employee | Increase if short excerpts miss the discussion's point | Can add unrelated or older detail. Dates identify when it was said. |
| Total employee-to-employee messages | Increase for more recent delegation awareness | More input cost; the total text ceiling still applies. |
| Messages per teammate | Increase for more detail from one teammate | Lower values limit any one peer's contribution; the total-message ceiling still applies. |
Start with defaults and change one setting at a time. Conversation controls show the current default and allowed range; choosing the displayed default restores that baseline. Failed saves display an error and preserve the saved value. The model-usage chart is read-only, not a routing control.
Reasoning Models
Some models stream a visible thinking process, shown as a "thinking" phase in the chat before the final answer. Reasoning capability is rated with stars (1 to 5) rather than a fixed on/off label, since models vary in how deep and how controllable that reasoning is.
For a model that streams reasoning, you can tune how hard it thinks with the Thinking Depth selector in the Employee Brain card: Auto, Off, Minimal, Light, Medium, or Deep. Deeper settings use more credits per message because of the additional thinking. For a model that does not stream reasoning, this selector is disabled.
How Models Affect Credits
Credit cost is set per model, not strictly per tier. A cheap Fast-tier model and a cheap Advanced-tier model can land close together, so check the exact rate shown next to each model before switching your whole workspace to it.
| What Changes | Impact |
|---|---|
| Cheaper model | Fewer credits per message, good for high-volume, everyday work |
| More capable model | More credits per message, reserve for work that truly needs it |
| Deeper thinking depth | Additional credits for the thinking phase, on models that support it |
Tips and Tricks
- Check the credit rate, not just the tier. Two models in different tiers can cost about the same; the number next to each model is the real signal
- Sort by Cheapest in the picker when you want the lowest-cost model that still fits your work
- Sort by Smartest to see the highest reasoning-rated models first, regardless of tier
- Use the Reasoning-only filter if the work needs a visible, step-by-step thinking process
Good to Know
- Changing the model here applies to every employee, immediately, on their next turn. There is no per-employee model setting in the interface today
- Model choices follow workspace access and any displayed cost ceiling. Locked choices explain the upgrade needed; see pricing.
- Free and Pending workspaces start on MiniMax M3.1 Flash. Paid plans start on Kimi K3.1 Pro, which is also used for Model Boost
- Workspaces using the previous MiniMax M3 or Kimi K3 defaults move to the corresponding newer model. Both previous models remain available in the picker
- Gemini 3.5 Flash Lite and GPT-6.1 Terra are available in the picker. Credits Saver settings are unchanged
- A Credits Saver toggle sits next to the model picker: when on, it automatically routes simple conversational turns to a cheaper model and keeps your chosen model for real work. Its own page shows tracked credits saved, the router cost for those same turns, and the net benefit.
Frequently Asked Questions
Q: Where do I set the AI model for my workspace? A: Go to Settings (gear icon) > Technical > AI Models. In the Employee Brain card, pick the model you want. It saves automatically.
Q: Can different employees use different models? A: Not from the interface today. The model you set on the AI Models Settings page applies to every employee in the workspace.
Q: What is the difference between tiers? A: Fast is built for simple, quick tasks. Standard balances cost and capability for everyday work. Advanced and Frontier handle complex reasoning and the hardest problems, generally at a higher credit cost, though the exact rate depends on the specific model.
Q: Do reasoning models cost more? A: A model that streams a visible thinking process uses more credits when set to a deeper Thinking Depth, because that extra thinking is billed like any other output.
Q: Why can't I change the model? A: Your workspace may have a model-cost ceiling. Choices above that limit are locked; use the upgrade prompt for full selection access. Choices within your permitted limit remain selectable.