OpenAI Releases New AI Models That Can Be Run On Laptops And GPUs

JAKARTA - OpenAI has announced two new AI models with open weight, gpt-oss-120b and gpt-oss-20b. These models are designed to be more accessible to developers due to their relatively light resource needs.

Unlike the ChatGPT model which is a closed model, gpt-oss-120b and gpt-oss-20b are "open-weight". This means developers can more freely use it without having to pay for API access.

These new models are light enough so they can be used without advanced hardware. gpt-oss-120b can run on one NVIDIA GPU, while smaller gpt-oss-20b can operate on ordinary laptops with 16GB RAM. This really helps developers who do not have access to expensive infrastructure.

According to OpenAI, gpt-oss-120b have capabilities that are almost equivalent to their o4-mini model in reasoning benchmarks, while gpt-oss-20b gives results similar to those of o3-mini.

If developers need greater processing power, these models can also send queries to the stronger AI model in the cloud.

Benchmark Test Results

In terms of performance, these models show impressive results. In Codeforces, gpt-oss-120b and gpt-oss-20b, each scored 2622 and 2516, ahead of DeepSek R1 models.

On the Humanity's Last Exam test which includes various subjects, their score is also quite good, namely 19% for gpt-oss-120b and 17.3% for gpt-oss-20b. Both scores are higher than competitors such as DeepSek and Qwen.

The release of these models is in line with the US government's push under Donald Trump's administration for AI developers in America to release more open-source technology. It also responds to the popularity of open models from Chinese companies such as DeepSek, Alibaba, and Moonshots.