Jais (Inception / G42)
UAE's open-weight Arabic LLM family, now Jais 2 at 70B params
Overview
Jais is an open-weight Arabic-English LLM family built by Abu Dhabi's Inception with MBZUAI and Cerebras; the Jais 2 generation (70B, Apache 2.0) targets developers and sovereign-AI teams needing an Arabic-first alternative to closed multilingual models.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 12 Jul 2026 and may be out of date or incomplete.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 Jais (Inception / G42)
- Open-weight model family, 590M to 70B params, Apache 2.0
- Jais 2 flagship at 70B and 8B, trained on an Arabic-first corpus
- Runs at roughly 2,000 tokens/sec on Cerebras hardware
- Free chat app at jaischat.ai
- Handles Modern Standard Arabic, regional dialects and code-switching
- Hugging Face weights for self-hosted deployment
ASEAN Perspective
Jais (Inception / G42) in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
Jais earns its reputation as the most credible open Arabic LLM on Hugging Face: an eight-size family from 590M to 70B params, all Apache 2.0, backed by MBZUAI's research bench and Cerebras compute. Jais 2 pushes further, running on Cerebras hardware at around 2,000 tokens a second and handling Modern Standard Arabic, regional dialects and code-switching better than earlier releases.
The caveat is in the fine print: independent benchmarks on dialectal and cultural tasks have shown general multilingual models like Qwen-7B beating the larger Jais-13B, and proverb or cultural-construct handling still lags. "Best Arabic model" doesn't mean best on every Arabic task. There's also no public API pricing listed anywhere — deployment means self-hosting the weights or going through Inception's enterprise sales. Worth it for teams building Arabic products who want open weights; skip it if you just need decent Arabic support inside a frontier multilingual model.
What people say
Inception traces back to the Inception Institute of Artificial Intelligence, founded in Abu Dhabi in 2017 as the region's first dedicated AI research organization, later folded into G42. The first Jais model — 13B parameters, trained on a purpose-built 116-billion-token Arabic corpus — launched in August 2023 through a three-way partnership between Inception, MBZUAI and Cerebras Systems. A 30B version followed in November 2023. The company was merged into Core42 in 2023, then relaunched as a standalone Inception in August 2024 and repositioned as a product-first developer through 2025.
Jais 2 is the current flagship: 20 models across 8 sizes (590M to 70B), all released under Apache 2.0 on Hugging Face, trained on what Inception describes as its richest Arabic-first dataset yet. The chat app at jaischat.ai runs on Cerebras hardware at roughly 2,000 tokens per second — genuinely fast — and is built to handle informal tone and code-switching between Arabic and English, not just formal Modern Standard Arabic.
Independent academic evaluation complicates the marketing claim of "world's most advanced Arabic LLM." A 2026 dialectal and cultural benchmarking study (AraDiCE) found the smaller, general-purpose Qwen-7B outperforming Jais-13B by roughly 10 points on overall dialectal and cultural accuracy, even though Jais led specifically on location recognition (67.6%). Jais and similar Arabic-native models also scored lower on proverb and culturally rich content across several 2026 papers (DialectalArabicMMLU, Absher). Jais remains a standard baseline cited across this research, which says something about its centrality even where it doesn't win outright.
There's no meaningful consumer review trail — no G2, no Capterra, no Reddit threads — because Jais isn't sold as a SaaS product. Its actual user base is researchers downloading weights from Hugging Face and enterprise teams in the Gulf building sovereign-AI deployments, not end users comparing chatbots. Jais also shows up as a supported model family inside NVIDIA's NeMo AutoModel framework, which gives enterprise MLOps teams a documented path to fine-tune and serve it alongside other open-weight models rather than treating it as a one-off integration.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Jais (Inception / G42)'s official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Jais (Inception / G42) 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 Jais (Inception / G42) directly →
Spotted something out of date? Suggest an update →
Jais (Inception / G42) in the news
Alternatives to Jais (Inception / G42)
More in LLMs & Chat
