AI Tiers & Models¶
This document is about the Tiers and other nuances we have in our AI Gateway offering.
AI Tiers¶
Here are the AI Tiers we offer. For what a tier costs, contact us at support@dropsolid.com or ask your Dropsolid contact person.
| AI PoC | AI Essentials | AI Pro | AI Enterprise |
|---|---|---|---|
| Ideal to kickstart AI innovation in your organization using Drupal as your distribution mechanism for AI. Prove to the organization it works. Monthly renewal after a 2-month initial commitment. | To run basic AI features (CKEditor, AI Translate, content suggestions, lifecycle automation) on your Drupal site through a certified platform solution. Yearly contract. | AI pro allows you to run powerful AI powered tools using Basic and Advanced models, including the Dropsolid Preferred models. | Enterprise Dropsolid AI allows you all the flexibility and sovereignty to deploy AI into your DXP. Ideal for teams that are in regulated industries |
| Allocated Space: 10GB | Allocated Space: 10GB | Allocated Space: 10GB | Allocated Space: 30GB |
| Gives access to: Trial any feature | Allows you to run:
|
Allows you to run:
|
Allows you to run:
|
AI Available Models¶
All tiers include a budget for tokens. You can pick any of the available models below.
To get the accurate pricing per model, please contact us at support@dropsolid.com.
Note: all these models are served from within the EU.
| Model | Model ID | Tier | Status |
|---|---|---|---|
| Mistral 3.1 Small | eu-mistral-small-cloud |
Basic | ACTIVE |
| Gemini 2.5 Flash | eu-gemini-2.5-flash-cloud |
Basic | ACTIVE |
| Claude 4.5 Haiku | eu-claude-4.5-haiku-cloud |
Basic | ACTIVE |
| Gemini 3.8 Flash | eu-gemini-3.8-flash-cloud |
Advanced | NEW |
| Claude Sonnet 5 | eu-claude-5-sonnet-cloud |
Advanced | NEW |
| Gemini 2.5 Pro | eu-gemini-2.5-pro-cloud |
Advanced | ACTIVE |
| Mistral Medium 3 | eu-mistral-medium-cloud |
Advanced | ACTIVE |
| Gemini 3.5 Flash | eu-gemini-3.5-flash-cloud |
Advanced | DEPRECATED Removed on 2026-10-25. Switch to Gemini 3.8 Flash. |
| Claude 4.6 Sonnet | eu-claude-4.6-sonnet-cloud |
Advanced | DEPRECATED Removed on 2026-10-25. Switch to Claude Sonnet 5. |
Dropsolid Preferred Models¶
Not sure which model to pick? Use one of our two preferred aliases instead of a specific model ID:
| Alias | Currently points to |
|---|---|
dropsolid-preferred-basic |
Mistral 3.1 Small |
dropsolid-preferred-advanced |
Gemini 3.8 Flash |
The advantage: you point your integration at the alias once, and we keep it pointed at a model that handles what the previous one did, at a good price for its tier. When the market moves, you get the better model without changing any code or configuration on your side.
How we pick: we review the preferred models every two months. A model has to be generally available, serverless, and processed within the EU, and it has to offer the best quality for its price in the tier. We only switch when a new model is clearly better, not marginally better.
Note
Because the model behind an alias can change, its output can change too. If you need identical behaviour over time (for example for evaluations or finely tuned prompts), call a specific model ID from the table above instead.
Image recognition support¶
All the models we offer - can understand images. The pricing and quality of image-recognition depends on the model and the image size. Images are just transformed into the input tokens, here's an example table:
| Image size | Tiles | Input tokens |
|---|---|---|
| ≤ 384×384px (any format) | 1 | 258 |
| 800×600px (typical photo) | ~2 tiles | ~516 |
| 1920×1080px (Full HD) | ~6 tiles | ~1,548 |
| 3840×2160px (4K) | ~20 tiles | ~5,160 |
Image generation support¶
We also offer Image Generation Models - these can be used as placeholders, article covers, or contextual suggestive images in both websites and marketing materials.
To get the accurate pricing per model, please contact us at support@dropsolid.com.
- Imagen 4 Fast
- Imagen 4 Standard
- Imagen 4 Ultra
Vector Store (Database)¶
Tiers Pro & Enterprise come with a Vector Store Database. This allows you (or, your developers) to create systems with dynamic memory, i.e. for Retrieval-Augmented-Generation (RAG) setups. It allows you to store embeddings and to find closest matches.
You will get 10GB of disk-space for your database, but it can be expanded later. Enterprise users get 30GB.
Langfuse (Observability)¶
Tiers Pro & Enterprise come with Langfuse integration. This gives you (or your developers) full observability over what's happening with your AI implementation — every interaction with your system is automatically logged in Langfuse. Tracing and memory data share the same disk-space limitation as your vector database (10GB for Pro, 30GB for Enterprise), so retention is bounded by storage rather than by time. The Enterprise tier additionally gets full feature access to Langfuse.
Custom Embedding Models¶
After making many analysis, we narrowed down our search to an embedding model that performs very well in Dutch in various domains: multilingual-e5-large
We're providing access to this model for free for all Tiers and it's a sovereign, self-hosted model (Dropsolid AI is hosting it). We're doing reviews every quarter and test other models, so potentially more models will appear in our offering - specialized on multilingual setups.
Chunking text for embeddings¶
The multilingual-e5-large model accepts at most 512 tokens per input. Longer chunks are rejected. To split your content safely, use the AI Gateway's chunking endpoint: POST /utils/chunkinizer. Send your raw text together with max_tokens and overlap, and it returns ready-to-embed chunks with their token counts. It uses the same tokenizer as the embedding model, so no returned chunk goes over the limit. It requires the same API key as your other AI Gateway calls.