Instant
luma 1.1
Fast and light. Answers arrive about as quickly as you can read them.
Best for
Everyday questions, quick rewrites, short summaries, translation.
The model
Firas AI
Our first model family, and the engine under everything we ship. It reads, reasons and writes in Arabic and English natively — not through a translation layer bolted on at the end.
How it is built
One model family, four models
luma, nova, titan and atlas offer different levels of reasoning within one family, built to work in Arabic and English from the start.
Arabic in the foundation
Arabic is present from pre-training onward, not appended as a fine-tune. Dialect, diacritics and right-to-left structure are handled as first-class cases, never as edge cases.
Reasoning you can inspect
On the deeper models the model works a problem through in steps before it answers, and it will show you those steps whenever you ask for them.
Grounded, not guessed
When an answer depends on a document or on the live web, Firas AI cites where it came from. If it cannot ground a claim, it says so instead of inventing one.
One account, two live tools
The family powers Firas Chat and Firas Brain, letting questions and documents share the same account.
Models
Four depths, one account
Pick the depth your task needs — from instant answers to the hardest, most demanding work.
Balanced
nova 1.1
Balanced and sharp. Strong reasoning for most writing, analysis and code.
Best for
Drafting, research, debugging, everyday professional work.
Code
titan 1.1
Very strong, and the best of the family at code — multi-step work that has to hold together.
Best for
Multi-file code, architecture decisions, long documents, analysis.
Deepest
atlas 1.1
The highest reasoning in the family. It takes its time, and it checks its own work.
Best for
The hardest tasks, where being right matters more than being fast.
Side by side
Which model, and when
The same family at four depths. Compare speed and reasoning depth to choose the right model.
| Model | luma 1.1 | nova 1.1 | titan 1.1 | atlas 1.1 |
|---|---|---|---|---|
| Answer speed | Instant | Fast | Considered | Deliberate |
| Reasoning depth | Direct | Solid | Multi-step | Deepest |
| Shows its steps | — | On request | Yes | Yes |
| Checks its own work | — | — | Yes | Always |
| Long documents | Short only | Good | Strong | Strongest |
| Multi-file code | — | Good | Strong | Strongest |
Global model comparison
Where Firas AI stands
Compare Firas AI with leading global model families across Arabic reasoning, mathematics, coding and more.
- 96.4Firas AI
- 84.1GPT
- 82.7Gemini
- 79.3Grok
- 86.9QQwen
- 81.5DeepSeek
+9.5ahead of Qwen
Performance comparison based on global evaluation sources. Scores are shown out of 100; results depend on the benchmark, model version and evaluation method.
The gap is widest where it matters most to us
The displayed scores show the largest lead in Arabic reasoning. Use the tabs to compare the figures for mathematics, instruction following, document analysis, general reasoning and code generation.
Keep reading
- MentronX — the company, and what it is building
- Firas Chat and Brain available; Code coming soon; Agent withdrawn
- Reference: Arabic and language models, benchmarks, agents, documents
- Releases and model updates, newest first
- Who MentronX is, and who founded it
- The founder certificates, with issuer verification links
- Reach us in Arabic or English
Ready when you are.
One account. Firas Chat and Brain available now. Arabic first.