OpenAI GPT-OSS open-weight models and NVIDIA Nemotron 3 now available on Amazon Bedrock in AWS GovCloud

Amazon Web Services has expanded its Bedrock model catalog in AWS GovCloud (US) to include two OpenAI open-weight models and NVIDIA's Nemotron 3 series — extending access to frontier-class AI to government agencies and regulated industries operating within US compliance boundaries.

What's new

The newly available models in us-gov-west-1 (with cross-region inference extending to us-gov-east-1) are:

OpenAI GPT-OSS models:

• gpt-oss-120b: A 120-billion-parameter open-weight model described as designed "for production, general-purpose, and high-reasoning use cases" with a 128K-token context window and up to 16K output tokens
• gpt-oss-20b: A 20-billion-parameter model "designed for lower latency and local or specialized use cases," also with a 128K context window

Both GPT-OSS models were released under the Apache 2.0 license by OpenAI in 2025 — among the company's first open-weight releases since GPT-2.

NVIDIA Nemotron 3:

• Nemotron 3 Super 120B: A mixture-of-experts model with 120 billion total parameters activating roughly 12 billion per token, delivering "up to 5 times higher throughput" versus dense equivalents
• Nemotron 3 Nano 30B variants: Offers "4 times higher throughput than the previous generation and reducing reasoning-token generation by up to 60 percent," with a 1M-token context window

All models run with "inference running entirely on AWS-operated infrastructure," keeping data within the GovCloud compliance boundary.

Context

AWS GovCloud is the commercial cloud region designed to host sensitive government workloads — FedRAMP High, ITAR, and CJIS among others. Until now, the Bedrock model catalog available in GovCloud lagged significantly behind commercial regions, forcing government teams to either accept smaller models or move workloads outside the compliance boundary.

The expansion reflects two parallel trends: AI model providers are pursuing government distribution, and the federal market is accelerating AI adoption across defense, intelligence, and civilian agencies. OpenAI's GPT-OSS models landing on Bedrock GovCloud is particularly notable because OpenAI itself does not operate government-cleared infrastructure — AWS becomes the compliant distribution layer for models agencies otherwise could not run in a regulated environment.

Why it matters

Putting capable open-weight models inside a FedRAMP boundary removes a persistent obstacle for government AI programs. Agencies previously had to choose between compliant-but-weak models or capable-but-non-compliant deployments. Having gpt-oss-120b — which benchmarks near OpenAI o4-mini on reasoning tasks — available within GovCloud means agencies can pursue more ambitious AI applications without stepping outside their security boundaries.

For NVIDIA, GovCloud availability extends Nemotron 3's enterprise positioning into the government vertical, where long-context agentic workflows (the 1M-token window) have clear applications in document analysis, intelligence synthesis, and regulatory review.