PulseChoosing an AI model
The difference between the standard and reasoning models, which to use when, and how to think about time and cost.
By default, chat and recipes choose an AI model automatically according to the scope and the question. On some plans you can also specify the standard or reasoning model yourself. This article covers what each is like, how to choose, and what you can specify.
Team plan Enterprise plan
Choosing a model manually is available on the Team and Enterprise plans. The free plan only has automatic selection. What automatic selection does, and which tasks suit which model, are the same on every plan.
Which AI models are available in Pulse, and how they are combined, may change as provision changes. This article describes them by function — “standard model” and “reasoning (thinking) model” — rather than by individual model name. For the options actually available to you, see the model selector in the app.
Automatic selection
Leave model selection on “automatic” and the model is chosen according to where you opened chat and what you asked. For most purposes, automatic is fine.
You would switch manually when you want a fuller answer and are willing to wait, or when you want the work finished quickly.
The standard model
The standard model is a general-purpose model for everyday work. It responds quickly and suits short questions and routine output.
It is well suited to:
- asking a short question about a meeting
- pulling take-aways out of a discussion
- summarising what was said, briefly
- drafting a short email or message
The reasoning (thinking) model
The reasoning model works through its thinking in steps before answering. It suits producing long documents and analysis that cross-references several sources. It takes longer to respond than the standard model.
With the reasoning model, a line such as “thought for N seconds” appears above the answer. Open it and you can see the order in which it reached its conclusion. This helps when you want to trace the basis for a judgement, or verify output before using it internally.
A line above the answer showing the thinking time, expanded to show the stages of its deliberation.
It is well suited to:
- comparing several meetings to draw out common themes or trends
- working out an approach from what several people have said
- setting out the advantages and concerns of each option
- checking the reasoning towards a conclusion, rather than the conclusion itself
What each plan can specify
| Plan | Model selection |
|---|---|
| Free plan | automatic only |
| Team plan | automatic, or choose the standard or reasoning model manually |
| Enterprise plan | automatic, or choose the standard or reasoning model manually. Some models cannot be chosen until an administrator enables them |
The plan names (free / Personal / Team / Enterprise) are provisional. Rewrite them to match UNCHAIN, Inc.'s actual plans. The source article does not say whether manual selection is available on the Personal plan, so please confirm and add a row to the table above. It also does not give per-plan limits on chat usage or reasoning model usage, or what happens when they are exceeded. Please add those in line with UNCHAIN, Inc.'s pricing. For plans and billing generally, see Plans and billing.
Choosing by task
The longer and more complex the output, the more the reasoning model suits. For short checks and fixed-format output, the standard model means less waiting.
| What you want to do | Model that suits |
|---|---|
| Summarise one meeting | standard model |
| Extract decisions and next actions | standard model |
| Write a short message based on a discussion | standard model |
| Find “how was that decided?” within one meeting | standard model |
| Draw out trends across several meetings | reasoning model |
| Build a proposed approach from what several people said | reasoning model |
| Compare advantages and concerns to produce a discussion document | reasoning model |
| Check the reasons and assumptions behind a proposal | reasoning model |
The same thinking applies when specifying a model for a recipe. How to specify it is in the creation steps in Recipes (saved instructions), and how to make the tone and format consistent through the instructions is in Writing effective recipes.
Processing time and usage
Because the reasoning model works through steps before answering, it takes longer than the standard model — tens of seconds is not unusual. Using the standard model when you need a quick check and the reasoning model when you are settling in to produce a document keeps frustration with waiting down.
In general, the more meetings involved and the longer the output requested, the more computation is needed. Selecting only the meetings you need before running something produces a faster and more focused result than leaving the scope set to the whole home screen while asking for a long document. How to narrow the scope is explained in Asking your notes questions (chat).
Administrator restrictions
On the Enterprise plan, some models are disabled by default. While disabled they cannot be chosen even though they appear in the model selector. To use one, an administrator enables it under Settings > Workspace > Data security.
Administrators have this as a separate setting because data handling can differ between models.
Before asking an administrator to enable one, consider the following.
- whether the task really requires that model (the choice between standard and reasoning may be enough)
- whether the meetings involved contain client information or sensitive personal information
- whether that model is covered by an arrangement under which the provider retains no logs
About sending data to external model providers
Pulse stores data in a Japanese region by default. Some processing for AI summarisation and transcription may, however, be delegated to providers overseas. In that case data including meeting content is sent to that provider, which raises the question of cross-border transfer of personal data.
UNCHAIN, Inc. selects only providers where it can confirm contractually that the data sent is not used to train models, where the retention period is stated, and which have arrangements allowing supervision as a processor.
This provides that where the handling of personal data is delegated in whole or in part, the delegating party must exercise necessary and appropriate supervision over the processor. On business plans, the contracting company is the delegating party and UNCHAIN, Inc. the processor. Where UNCHAIN, Inc. uses an external model provider, that is a sub-delegation and subject to equivalent supervision.
This governs providing personal data to a third party in a foreign country. It applies even where provision accompanies delegation. UNCHAIN, Inc. provides data only to providers that have put in place arrangements meeting the standards set by Personal Information Protection Commission rules, and periodically confirms the measures those providers implement and the legal regime of the destination country.
For some models, an arrangement under which the provider retains no data (zero data retention) does not apply, and input and output logs may be retained for a period for purposes such as abuse monitoring. Whether your organisation uses such a model is for an administrator to decide through the setting above.
Whether data is used to improve Pulse's own AI models is a separate setting. On the free and Personal plans it is on by default and can be switched off at any time under Settings → Preferences → Data and sharing. On the Team and Enterprise plans it is not used for training. For details see AI models and training data, and for storage location and cross-border transfer generally see Security and privacy FAQs.
Which models fall outside zero data retention, the provider names, log retention periods and the countries where data is processed should be written to match UNCHAIN, Inc.'s contracts with each provider. The information provided under Article 28 of the Act on the Protection of Personal Information must be consistent with the Privacy policy.
Frequently asked questions
Q. I do not know which model to choose
A. Leave the setting on “automatic”. The model is chosen according to the scope and the question. It is simplest to switch to the reasoning model manually only when step-by-step deliberation is needed, or when you want to see the thinking.
Q. Automatic is the only option available
A. The free plan only has automatic selection. To choose a model manually you need to move to a paid plan from Settings > Billing. For what the plans include, see Plans and billing.
Q. A model in the selector cannot be selected
A. On the Enterprise plan, some models are disabled by default. Ask an administrator to enable it under Settings > Workspace > Data security. On the free plan, you need to move to a paid plan.
Q. The reasoning model is slow to answer
A. Because the reasoning model works through steps before answering, it takes longer than the standard model. For a short check, switch to the standard model. Narrowing down the meetings involved also reduces the time.
Q. The same instructions produce different tone with different models
A. Models differ in their output tendencies. To keep tone and format consistent, state the polite or plain form, the order of headings, and the length per item explicitly in the instructions. How to word them is explained in Writing effective recipes.