SILMA AI
Dialect-aware Arabic and English TTS for voice agents, plus open Arabic models
Overview
SILMA AI is a voice AI lab selling SILMA TTS v2.0, a low-latency text-to-speech API in English, Modern Standard Arabic and the Saudi Najdi dialect, built for voice-agent platforms. It also publishes open models on Hugging Face: the SILMA-9B and Kashif 2B Arabic LLMs, Arabic embeddings and the 150M-parameter SILMA TTS v1.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete.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.
- 5 minutes, 5,000 characters
- No concurrency
- Voice cloning and pronunciation control
- Access to all TTS models
- 100 minutes (100,000 characters), $0.1 a minute
- 2 concurrent requests
- Voice cloning and pronunciation control
- 1,000 minutes (1,000,000 characters), $0.05 a minute
- 2 concurrent requests
- Voice cloning and pronunciation control
- 10,000 minutes (10,000,000 characters), $0.029 a minute
- 4 concurrent requests
- Voice cloning and pronunciation control
- 30,000 minutes (30,000,000 characters), $0.020 a minute
- 8 concurrent requests
- Voice cloning and pronunciation control
- Up to 20 concurrent requests
What you can produce with SILMA AI
- Streaming Arabic and English speech from the SILMA TTS v2.0 API over WebSockets, SSE or REST
- Saudi Najdi dialect and Modern Standard Arabic voices with code-switching in one pipeline
- Custom voices cloned from a 10-second sample, under a written-consent rule
- Pronunciation dictionaries and speed and variation controls
- Pipecat and LiveKit plugins for voice-agent pipelines
- Open-weight Arabic LLMs (SILMA-9B, Kashif 2B) and the Apache-licensed SILMA TTS v1
- The Arabic Broad Leaderboard and an Arabic TTS benchmark for comparing models
SILMA AI is now a focused Arabic voice vendor. TTS v2.0 covers Modern Standard Arabic, Saudi Najdi and English with code-switching, streams at a claimed 170 ms to first audio, and costs from $10 a month down to $0.02 a minute at volume, with monthly billing only and a free trial credit. Voice cloning sits behind a clear consent rule.
The open releases remain useful, including an Apache-licensed 150M-parameter TTS model, but the Arabic LLMs have not been updated since mid-2025. The company publishes little about who runs it, so check support terms before building production voice agents on it.
What we found
SILMA AI has changed what it is. It first appeared in August 2024 as the maker of SILMA-9B-Instruct, an open-weights Arabic model built on Google's Gemma 2, followed by Kashif 2B for retrieval-augmented question answering and a pair of Arabic embedding models. The 9B model card says it was the top-ranked open-weights Arabic LLM until February 2025; that is SILMA's own claim, though the benchmark scores it cites come from the public Open Arabic LLM Leaderboard. Today the homepage describes the company as a voice AI lab building low-cost, dialect-aware speech models for companies that make voice agents.
The paid product is SILMA TTS v2.0, a text-to-speech API covering English, Modern Standard Arabic and the Saudi Najdi dialect, with code-switching between them. SILMA quotes about 170 ms to first audio on its servers and 280 ms end to end in the EU, voice cloning from a 10-second sample, pronunciation dictionaries, speed and variation controls, and WebSocket, SSE and REST streaming. Repeated requests are cached and charged at half price, and there are Pipecat and LiveKit plugins. The terms make voice cloning conditional on the user's own voice or documented written consent.
Pricing is published and monthly only. A free account gets one-time credits of 5 minutes or 5,000 characters. Starter costs $10 a month for 100 minutes, Pro $50 for 1,000, Growth $299 for 10,000 and Scale $599 for 30,000, which works out to $0.02 a minute. An Enterprise plan with up to 20 concurrent requests is mentioned without a price.
The open work continues alongside: SILMA TTS v1, a 150-million-parameter bilingual model on the F5-TTS architecture with Apache 2.0 weights, the Arabic Broad Leaderboard and an Arabic TTS benchmark. The operator is SILMA.AI LLC, a Delaware company; it does not name its founders or location. For teams building Arabic voice agents the pricing is easy to test. For Arabic text generation the LLMs are now older and unchanged since mid-2025.
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 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
