We use cookies to personalize content and to analyze our traffic. Please decide if you are willing to accept cookies from our website.
Flash Findings

OpenAI Opens Up: What You Should Know About GPT-OSS

Mon., 25. August 2025 | 1 min read

Quick Take

OpenAI has, for the first time since GPT-2 in 2019, unleashed two open-weight reasoning models, gpt-oss-120b and gpt-oss-20b. This is a strategic shift driven by mounting competition from DeepSeek and other open-source players. With more open-weight models available, IT leaders gain leverage and flexibility. This growing choice pressures vendors to accelerate innovation and offer better open (open-weight and open-source) models.

Why You Should Care

  1. Engineered for reasoning & efficiency. The new models are built using a mixture-of-experts (MoE) transformer architecture with chain-of-thought reasoning, excelling in benchmarks for mathematics, coding, health, and tool use.
  2. Works with accessible hardware. The 117B-parameter, gpt-oss-120b, runs on a GPU with 80 GB of RAM, while the lighter 21B-parameter, gpt-oss-20b, runs on a GPU with 16 GB of RAM, like a consumer-grade laptop.
  3. Low cost, high customizability, and high governance. You get full model inspection, inference locally behind your firewall, and flexibility to fine-tune or embed these models into secure, regulated workflows via platforms like AWS, Azure, Databricks, and Hugging Face.

What You Should Do Next

Now is the time to take a closer look at OpenAI’s open models and other open alternatives.

Get Started

  • Download gpt-oss-20b and kick off pilot programs using gpt-oss-20b on developers' workstation to benchmark its reasoning, latency, memory footprint and chain-of-thought features.
  • Recalculate your AI spend model by comparing open-weight deployment costs versus closed-API fees and lock-in risks. Model your TCO impact by comparing infrastructure, licensing, and maintenance costs against currently used proprietary or open models. 
  • Engage your security and compliance teams early. Open-weight models are perfect for affordability, but also require robust governance to manage jailbreak risks and data confidentiality. Embed these models into your AI governance framework, if you choose to use them. Govern them properly so they do not create compliance headaches down the road.

Learn More @ Tactive