SILMA AI
Open Arabic LLMs and dialect-aware TTS, free-weighted on Hugging Face
Overview
SILMA AI builds open-weights Arabic LLMs and a bilingual TTS model with instant voice cloning, plus Saudi Najdi-dialect speech, for developers and enterprises who need Arabic-native language and voice tools rather than translated Western ones.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 4 Sep 2026 and may be out of date or incomplete.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 SILMA AI
- SILMA-9B-Instruct: open-weights Arabic LLM built on Gemma2 (Gemma license)
- SILMA TTS v1: 150M-param bilingual Arabic/English voice model with instant voice cloning
- Dedicated Saudi Najdi-dialect TTS model
- Open Arabic LLM Leaderboard (ABL) benchmark suite
- Custom Arabic model development and consulting
- Open distribution via Hugging Face, GitHub and Ollama
SILMA AI's pitch is narrow and well-executed: Arabic speech and text, done properly, instead of bolted onto a Western model as an afterthought. SILMA TTS v1 is a 150-million-parameter bilingual Arabic/English voice model built on F5-TTS with instant voice cloning and a dedicated Najdi-dialect Saudi variant — genuinely useful if you need Arabic speech synthesis and don't want to fight accent artifacts. SILMA-9B-Instruct, built on Google's Gemma2, held the top spot on the Open Arabic LLM Leaderboard among open-weights models until February 2025 despite being a fraction of the size of rivals.
Everything ships open — MIT-licensed code, Hugging Face weights, an Ollama listing — so you can evaluate it yourself for free before any consulting engagement. The tradeoff is scale: this is a small team (GitHub follower and star counts are in the dozens, not thousands), so support and roadmap pace will feel more like an open-source project than an enterprise vendor.
What people say
There's no consumer review trail for SILMA AI — no G2, no app-store ratings — because the audience is developers and enterprises evaluating Arabic-language models, and the evidence that matters lives on Hugging Face and GitHub instead.
SILMA-9B-Instruct, built on Google's Gemma2 architecture, pulls around 2,980 downloads a month on Hugging Face and scores 78.19% on Arabic ARC Challenge, 78.89% on ACVA and 86% on Arabic ARC Easy on the Open Arabic LLM Leaderboard. SILMA's own documentation describes it as the top-ranked open-weights Arabic LLM until February 2025, "surpassing models over seven times larger" — a claim worth treating as a vendor benchmark rather than an independent audit, but the underlying leaderboard numbers are public and checkable.
On the speech side, SILMA TTS v1 is a 150M-parameter bilingual Arabic/English model built on the F5-TTS diffusion architecture, released under a permissive license with instant voice cloning and pronunciation control, plus a dedicated model for the Saudi Najdi dialect. The company publishes explicit voice-cloning consent rules — cloning requires documented permission from the speaker and bars use for deepfakes or fraud — which is more governance than most TTS vendors bother to state up front.
The GitHub organization (SILMA-AI) has three public repos: silma-tts (26 stars, 6 forks, MIT license), an Arabic-AI-ecosystem tracker, and a resources repo, all actively updated as recently as May 2026. That's a small footprint — this reads as a focused research-and-consulting shop rather than a company with thousands of production deployments — but for a team building open Arabic-native models, the work is real and independently verifiable rather than marketing copy.
Beyond the flagship LLM and TTS releases, SILMA also runs a public Arabic LLM leaderboard (ABL) that ranks other vendors' models on the same benchmark suite it uses for its own, and it lists premium consulting and custom model development as a separate service line for enterprises that need something beyond the open releases. None of that shows up as a star rating anywhere, but combined with the download and benchmark numbers it's a reasonable proxy for a working, actively maintained business rather than an abandoned research project.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including SILMA AI's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with SILMA AI 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 SILMA AI directly →
Spotted something out of date? Suggest an update →
Alternatives to SILMA AI
More in Video & Audio
