OpenAI has launched two new artificial intelligence (AI) language models that are free to use and can run directly on consumer computers.
The larger model, called gpt-oss-120b, works with one 80GB graphics card. A smaller version, gpt-oss-20b, needs 16GB of memory.
Both models support long conversations or inputs, up to 128,000 tokens, which matches the limit set by OpenAI’s GPT-4o.
Did you know?
Subscribe – We publish new crypto explainer videos every week!
Toobit Tutorial For Beginners (FULL Animated 2025 Guide)
In an August 5 announcement, OpenAI stated that the new models are designed to perform well in real-world use while still being easy to run on common devices.
These models are available under the Apache 2.0 license, which allows both personal and commercial use. OpenAI said they perform about as well as its o4-mini model when tested on tasks that require reasoning.
Instead of using all parameters at once, the models use a “mixture-of-experts” design, which activates only part of the full system for each task.
Developers can choose how much processing power to use by selecting from three levels of reasoning effort: low, medium, or high. These settings are easy to adjust with a short instruction in the system message.
OpenAI trained the models using techniques from earlier systems like o3, including reinforcement learning and fine-tuning. After training, they went through an extra process to make them safer and more reliable.
The models were also trained to avoid harmful answers and resist tricks that try to bypass safety controls.
Recently, OpenAI announced plans to use a new data center in northern Norway to support its AI operations in Europe. What did the company say? Read the full story.
