Jais 3: six open Arabic-native models (0.9B-375B) with speech and agents
MBZUAI's Institute of Foundation Models and Inception42 launched Jais 3: six open Apache-2.0 Arabic-native models (0.9B-375B, 512K context) with dialect-aware Jais Speech, built on K2 Horizon with audited Arabic benchmarks released publicly.
Why it matters
First adaptation of an open foundation to Arabic with full research traceability — five training checkpoints per model plus a public 24-benchmark audit suite — at 7B jumping from 56.4 to 79.6 Arabic score vs base.
What you could build with it
A Middle East fintech startup could build an Arabic voice-first customer-service agent on the 7B Jais 3 model plus Jais Speech, running it in-country to satisfy sovereign data requirements.
Does it hold up?
Models available on Hugging Face with vLLM and SGLang support and publicly released evaluation code; third-party evaluations of the weights have not appeared yet.
Built with Jais 3
- Aletihad: MBZUAI, Inception42 launch Jais 3 Arabic AI modelsblog · WAM-wire announcement of the Jais 3 fleet: six open models from 0.9B to 375B paired with Jais Speech.
- Gulf Time: MBZUAI, Inception42 launch Jais 3 Arabic AI modelsblog · Gulf press coverage confirming the six-model fleet, 512K context, and the public benchmark release.
Learn more
First spotted on article: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74StepFun Step-5-Preview — open BF16 weights releasedStepFun published the full BF16 weights for its flagship Step-5-Preview agentic model (600B-parameter sparse MoE, 27B…models · JEV 0.73ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71JetBrains Mellum 2.1: 12B MoE coding model trained for agentsJetBrains released Mellum 2.1, a 12B MoE model (2.5B active parameters, 128K context) under Apache 2.0, trained mainly…models · JEV 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.