Hi all!!

We're excited to launch OpenAI's new pair of Open Models on GroqCloud! gpt-oss marks OpenAI's first open-weight language models since GPT-2 These models deliver strong real-world performance at low cost, trained using reinforcement learning and techniques informed by OpenAI's most advanced internal models, including o3. And on Groq they can run at:

• gpt-oss-120B (117B params, 5.1B active) running at 500+ t/s,
• gpt-oss-20B (21B params, 3.6B active) running at 1000+ t/s,

Both models are available NOW to all users, including free tier and developer tier What makes gpt-oss special:

• Achieves near-parity with o4-mini (120B) and o3-mini (20B) on core reasoning benchmarks,
• Strong tool use capabilities and few-shot function calling,
• Full chain-of-thought reasoning with adjustable reasoning effort (low/medium/high),
• Mixture-of-Experts architecture using alternating dense and locally banded sparse attention,
• 128K native context length with 32K max output tokens,

But wait, there's more! 
We're also launching Built-in Tools for agentic workflows:

• Web Search - Real-time internet browsing,
• Python Code Execution - Run code directly in your workflows,
• For a limited time, tool use is completely FREE!

But wait, there's EVEN MORE! 
Introducing the Responses API (beta) - fully compatible with OpenAI's Responses API!

The Responses API is designed for agentic workflows with exceptional instruction following, tool use, and reasoning capabilities. The Responses API supports both regular tool use AND the new built-in browser search and code execution tools (exclusive to gpt-oss models for now)!

Now it's YOUR turn to show me what you've got!

Build something cool with these models, make a demo, post it on socials and in the community (or Discord!), and I'll give $50 credits to my favorites!

Read more about all our launches here: https://console.groq.com/docs