OpenAI Went Open Source and It’s a Big Win for AMD (and You)

This week, OpenAI broke character and dropped something no one expected: two open-weight language models under the Apache 2.0 license.

Meet gpt-oss-120B and gpt-oss-20B, OpenAI’s fully transparent, fully local, and shockingly efficient models you can run yourself. No APIs. No limits. Just weights.

But here’s the kicker: they’re optimized to run on AMD hardware, and TensorWave has AMD hardware at scale.

Let’s talk about why this changes the game.

What Just Happened?

OpenAI released two language models:

• gpt-oss-120B (≈117B params): Matches o4-mini (Claude 3 Haiku–level reasoning) and runs on a single 80GB GPU.
• gpt-oss-20B (≈21B params): Matches o3-mini (GPT-3.5–like performance) and runs on machines with just 16GB of RAM.

These aren’t just “research-only” models. You can:

• Fine-tune them
• Embed them locally
• Deploy them in production

Why It Matters

1. Freedom to innovateYou control the weights. You control the stack. Build, test, deploy—without waiting on some company to approve your use case.
2. Lower infra costsYou don’t need a cluster of 8 GPUs to run the 120B model. You need one Instinct MI300X or one H100. (Guess which one has 2x the VRAM?)
3. Performance at your fingertipsThe smaller 20B model runs on devices with 16GB of RAM. Think laptops, desktops, even edge boxes.

This is the democratization of high-quality LLMs, and OpenAI just handed everyone the keys.

Yes, It Runs on AMD

From day one, OpenAI’s gpt-oss models run great on AMD:

• On desktop: Ryzen™ AI Max+ laptops can run gpt-oss-120B with 128GB RAM
• On edge: Radeon™ 9070 XT (16GB) supports fast inference for gpt-oss-20B
• In the cloud: AMD Instinct™ MI300X, MI325X, and MI355X scream through inference and training workloads

And if you want to scale that performance?

It Runs Even Better on TensorWave

TensorWave is the AMD GPU cloud purpose-built for performance. And that means:

• You get 256GB of VRAM per GPU—perfect for 120B-sized models without model parallel hell
• You get ROCm-optimized infrastructure that plays nice with Hugging Face, PyTorch, and DeepSpeed
• You get scale without compromise—dedicated clusters, elastic capacity, zero vendor lock-in

Whether you’re experimenting with OSS models or deploying them in production—we’ve got the infrastructure dialed in.

Want to Try It?

Spin up an AMD Instinct Series node (we have MI300X, MI325X available now) in the TensorWave cloud. Drop in gpt-oss-20B or 120B. See what kind of magic happens when open source meets open hardware.

No lock-in. No nonsense. Just raw AI performance.

👉 Reserve Your AMD Node Now

About TensorWave

TensorWave is the AMD GPU cloud purpose-built for performance. Powered exclusively by Instinct™ Series GPUs, we deliver high-bandwidth, memory-optimized infrastructure that scales with your most demanding models—training or inference.

Ready to get started? Connect with a Sales Engineer.