US AI infrastructure specialist Groq and Saudi Arabia’s Public Investment Fund–owned HUMAIN have rolled out OpenAI’s two new open-weight language models, gpt-oss-120B and gpt-oss-20B, on GroqCloud with immediate global availability, including full regional access from Groq’s Dammam data centre.

The release gives developers in Saudi Arabia and across the Middle East day-zero access to OpenAI’s first open-weight reasoning models since GPT-2 more than five years ago. Both models arrive with full 128K context length, integrated server-side tools, and local support through HUMAIN.

GroqCloud is offering gpt-oss-120B from $0.15 per million input tokens and $0.75 per million output tokens, with the smaller gpt-oss-20B priced at $0.10 and $0.50 respectively. The larger model can run on a single Nvidia GPU, while the lighter 20B version can operate on consumer laptops with 16GB memory. Performance benchmarks show speeds of over 500 tokens per second for gpt-oss-120B and 1,000 tokens per second for gpt-oss-20B.

HUMAIN, launched earlier this year as Saudi Arabia’s national AI company, is providing in-Kingdom support, billing in riyals, and compliance-ready data hosting. This means developers can start prototyping gpt-oss-20B locally, even on laptops, and scale to production workloads on Groq’s Dammam-based infrastructure without waiting for phased rollouts or competing for scarce GPUs.

For enterprises, GroqCloud’s regional presence offers low-latency deployment and the option to fine-tune gpt-oss-120B on private datasets for applications such as RAG pipelines, voice agents, and customer support chatbots.

Technically, the gpt-oss models mark OpenAI’s re-entry into the open-source arena, using a mixture-of-experts architecture that activates just 5.1 billion parameters per token despite the 117-billion-parameter scale of the 120B model. Both were trained with reinforcement learning methods from OpenAI’s most advanced internal systems, including o3, and achieve near-parity with proprietary models on reasoning benchmarks. Released under the permissive Apache 2.0 licence, they are also freely available on Hugging Face for broader experimentation.

By pairing these models with Groq’s global infrastructure footprint, which spans North America, Europe, and the Middle East and is already used by more than 1.9 million developers, the partnership positions Saudi Arabia as an early access hub for the latest generation of open-weight AI systems.