OpenAI releases free AI models for first time since 2019

For the first time since 2019, OpenAI released two free AI models for download. One will even run on your laptop. Here's all you need to know.

After several delays, OpenAI on Tuesday released two new AI models. And for the first time since GPT-2 in 2019, you can run them on your own systems.

What did OpenAI release?

The two AI models are:

• gpt-oss-120b.
• gpt-oss-20b.

The first can run on a single Nvidia AI GPU with 80GB of memory, and the smaller one will run on a consumer-grade system with at least 16GB of memory.

• Notably, the two “open-weight” AI models are free for developers to use and customise. It's also released under the permissive Apache 2.0 license, which allows for monetisation without having to pay OpenAI.
• Both are reasoning models that are text-only. OpenAI says the larger gpt-oss-120b has comparable performance to OpenAI's o4-mini, while the smaller gpt-oss-20b is comparable to o3-mini.

What changed?

So why the sudden change of heart? I reckon you can thank DeepSeek.

Days after the open-source DeepSeek R1 was unveiled in January, CEO Sam Altman publicly said he needed to figure out "a different open-source strategy."

This week's release is Sam's response.

For context, DeepSeek R1 was released under MIT License, and its model weights and code are both freely available for download and commercial use.

Meta's Llama AI models are also open-weight, even though it persists in mislabelling it as "open source." Moreover, it imposes various restrictions on use that were only gradually being loosened.

What does it mean?

OpenAI says its new open-weight reasoning models can write code and are capable of using tools such as web search.

While OpenAI did not release benchmarks, anecdotal evidence from experts indicate that it offers state-of-the-art performance among other open models.

• In theory, developers can run the smaller AI model on their machines to operate agents that call on APIs or access resources via the MCP protocol.
• Amazon has announced that OpenAI's new models are already available on its cloud-hosted AI platform, Amazon Bedrock.

For now, the AI community is also eagerly awaiting the release of DeepSeek R2, the next iteration of DeepSeek's own reasoning model.