Model catalogue
One dependable model first. More will follow.
Qabas currently offers one text model. We would rather document one working service honestly than show a long catalogue that is not ready.
Live
Qwen 2.5 7B Instruct
qwen-2.5-7b- Best for
- Everyday text, chat, summaries and bilingual prototypes
- Languages
- Multilingual, including Arabic and English
- Current context
- 4,096 tokens for the request and reply together
- Input
- Text
- Output
- Text
- Underlying model
Qwen/Qwen2.5-7B-Instruct- Company origin
- UAE-founded
- Data residency
- Verified regional statement not published for this preview
Why Qabas starts with Qwen 2.5 7B
It gives the first Qabas service a practical balance:
- It is small enough to run on a dedicated single-GPU service.
- It can work across Arabic and English without a separate endpoint.
- It uses an open model that Qabas can host and operate directly.
- Its size makes it suitable for testing the gateway, keys, metering and reliability before adding larger models.
How the catalogue will grow
Future models will keep the same API address and key format. Each model page will state its input types, context length, price, region, known limits and benchmark method.
Planned categorySmall fast chat
For high-volume support and extraction.
Planned categoryLarge general model
For harder writing and reasoning tasks.
Planned categoryVision
For text plus images when supported.