AI models
9brains offers a selection of AI models from various providers. Administrators can determine in the model settings which models are available in the workspace.
In the model selector you see a cost category (€, €€ or €€€) next to each model. It indicates how much the model draws from your monthly quota. Premium models (€€€) deliver the best results but consume more quota per request. More on this under Budget & quota.
Model comparison
Section titled “Model comparison”Artificial Analysis Intelligence Index v4.3 (as of 11 Sep 2026), higher is better
Faded and italic: no cost measurement of our own yet, the value is extrapolated from the list price. Only the cost is affected, not the intelligence score.
The stated costs apply to a demanding request with several follow-up questions and repeated tool use (such as a knowledge-base search and a short calculation). A simple single question without tools is about five to ten times cheaper depending on the model. All model costs are included in your monthly quota and are not billed on top, the values serve only to compare models.
Chat models
Section titled “Chat models”| Model | Location | GDPR-compliant? | Unlimited use* |
|---|---|---|---|
| OpenAI | |||
| GPT-6 Astra | Global (DPF certified) | Yes | |
| GPT-6.1 Sol | Paris, France | Yes | |
| GPT-6 Luna | Paris, France | Yes | Yes |
| GPT-5.6 Terra | Paris, France | Yes | |
| Anthropic | |||
| Claude Fable 5.1 | USA (DPF certified) | Yes | |
| Claude Opus 5.5 | EU | Yes | |
| Claude Opus 4.8 | EU | Yes | |
| Claude Sonnet 5.5 | EU | Yes | |
| Claude Sonnet 5 | EU | Yes | |
| Claude Haiku 5.5 | EU | Yes | |
| Gemini 4 Argon | EU | Yes | |
| Gemini 3.8 Flash | EU | Yes | |
| Z.ai | |||
| GLM 5.3 | EU | Yes | |
| GLM 5.3 Flash | EU | Yes | |
| Qwen (Alibaba) | |||
| Qwen3.8 Max | EU | Yes | |
| Qwen3.8 Flash Next | EU | Yes | |
| Moonshot AI | |||
| Kimi K3 | Paris, France | Yes | |
| MiniMax | |||
| MiniMax M3.1 Flash | EU | Yes | |
| DeepSeek | |||
| DeepSeek V4.1 Flash | EU | Yes | |
| xAI | |||
| Grok 4.3 | USA (DPF certified) | Yes | |
| Mistral AI | |||
| Mistral Large 4 | Paris, France | Yes | |
* Unlimited use: with a paid license you can keep using this model in the chat even when your quota is used up. During the trial, the daily limit still applies to it. Autonomous agent runs do not count against the quota, they are paid from credit and pause when it runs out.
Image models
Section titled “Image models”| Model | Location | GDPR-compliant? | Rating (create / edit) |
|---|---|---|---|
| Nano Banana Pro | USA (DPF certified) | Yes | 1102 / 1099 |
| Nano Banana 2.1 | USA (DPF certified) | Yes | — / — |
| Nano Banana 2 | EU | Yes | 1125 / 1108 |
| Black Forest Labs | |||
| FLUX.2 Pro | Paris, France | Yes | 1004 / — |
| FLUX.2 Flex | Global (DPF certified) | Yes | 1028 / 1006 |
| OpenAI | |||
| GPT Image 2 | Global (DPF certified) | Yes | 1172 / 1122 |
| GPT Image 2.5 Flare | Global (DPF certified) | Yes | 1191 / 1162 |
| GPT Image 2.5 Sunburst | Global (DPF certified) | Yes | 1197 / 1182 |
Rating: Elo score from blind comparisons, in which people pick the better of two images for the same request without knowing the models. Higher is better, a dash means not rated or, for editing, not supported. Source: Artificial Analysis
Classification models
Section titled “Classification models”These models do not write answers. They sort texts into fixed categories and say how sure they are: Which department is responsible? How urgent is it? Is this a complaint? This takes a fraction of a second and is much cheaper than a chat model. That makes them a good fit for large amounts, for example sorting a mailbox, tickets or a folder full of documents.
You do not call them directly. Simply ask the chat or an agent, for example “Sort today’s emails by department and flag the urgent ones”. The AI agrees the categories with you, sends the texts to the classification model and shows you the result. Unclear cases come back as a list for you to check. The AI only deletes or moves emails if you explicitly want it to.
What these models cannot do: count, calculate or compare dates. The AI therefore checks things like “more than three reminders” or “older than 30 days” itself beforehand.
Classification models are available from the Business plan up. Clef runs in the EU and does not store requests; it can be used in every data privacy mode and is the one used unless something else is set. Decision-1 by Microsoft also runs in the EU and can be used in every data privacy mode. As with the GPT models, Microsoft may store requests for a limited time to detect abuse, but does not train on them. Jev runs in the USA, and the provider may store requests. That is why it can only be used in the “Global” data privacy mode and is meant for non-sensitive data only. More on this under Data privacy.
| Model | Location | GDPR-compliant? |
|---|---|---|
| TypeSafe | ||
| Jev 1.13 | USA | — |
| Cloudflare | ||
| Clef | EU (multiple locations) | Yes |
| Microsoft | ||
| Decision-1 | Paris, France | Yes |