Fanar
Qatar's free, open Arabic-first AI models and chat from QCRI
Overview
Fanar is the Qatar Computing Research Institute's Arabic-first AI platform: Fanar-2-27B-Instruct with native Arabic reasoning and tool use, plus Sadiq, Oryx, Aura, Shaheen and Diwan models for Islamic Q&A, images, speech, translation and poetry. Free chat and apps, Apache-2.0 weights and a free rate-limited API.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 1 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
What you can produce with Fanar
- Fanar-2-27B-Instruct Arabic-first language model
- Sadiq Islamic question answering
- Oryx image generation and understanding
- Aura speech and Shaheen translation models
- Diwan Arabic poetry model
- Free chat, apps and rate-limited API
Fanar is the most complete Arabic-first AI offering from a public research institute: a capable 27B model with native Arabic reasoning, specialist models for Islamic questions, speech, images, translation and poetry, and everything free, with Apache-2.0 weights. Its benchmark wins are QCRI's own and larger models still score higher, and Fanar 3.0 is due in December 2026. An excellent choice for Arabic-language applications and research; outside Arabic, general-purpose models are stronger.
What we found
Fanar is a family of Arabic-first AI models and services from the Qatar Computing Research Institute at Hamad Bin Khalifa University, part of Qatar Foundation, supported by Qatar's Ministry of Communications and Information Technology. The first version, a 9-billion-parameter model, launched in December 2024; Fanar 2.0 was announced in December 2025 and its weights were published in March 2026. The main model, Fanar-2-27B-Instruct, is built on Google's Gemma 3 27B base, with a 32,768-token context, a thinking mode that reasons natively in Arabic, and tool calling, and was trained on 256 H100 GPUs using about 120 billion high-quality tokens, by its paper's count. Alongside it are Sadiq, for Islamic questions and answers, which QCRI says has handled millions of queries on IslamWeb and IslamOnline; Oryx for image generation and understanding; Aura for speech; Shaheen for translation; and Diwan for Arabic poetry. QCRI says every component was designed and operated entirely at QCRI.
QCRI's own evaluations show Fanar 2.0 improving on the first version, with ArabicMMLU rising from 67.35 to 74.67 and GSM8K from 83.02 to 93.70, and leading models of its size on several Arabic benchmarks, though its paper acknowledges that larger models such as Jais-2-70B and Llama 3.3 70B score higher. These are QCRI's results rather than independent tests. Its preference training used about 280,000 pairs, including feedback from users of the live service. QCRI says work has begun on Fanar 3.0, a mixture-of-experts model trained from scratch, for release in December 2026.
Fanar is free. There is a web chat at chat.fanar.qa and iOS and Android apps, the model weights are on Hugging Face under the Apache 2.0 licence, and developers can request free API access, limited to 50 requests a minute for the language models.
It is aimed at Arabic speakers and organisations in the Gulf and wider Arab world, and has little specific relevance to Southeast Asia.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Fanar's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Fanar unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Fanar directly →
Spotted something out of date? Suggest an update →
Alternatives to Fanar
More in LLMs & Chat
