gpt-oss-120b
OpenAI's gpt-oss-120b is a 117-billion-parameter, open-weight Mixture-of-Experts model optimized for complex reasoning and agentic workflows.
- 117B-parameter Mixture-of-Experts (MoE) architecture
- Text-only design optimized for reasoning and agentic tasks
- Released under the Apache 2.0 open-source license
- Locally deployable alternative to OpenAI's closed API models
Identity
gpt-oss-120b is a 117-billion-parameter open-weight model developed by OpenAI and released on August 5, 2025 (model profile). It utilizes a Mixture-of-Experts (MoE) architecture and is distributed under the Apache 2.0 license (model profile).
What it is
The model is a text-only system designed for complex reasoning and agentic workflows (model profile). It features a 131,072-token context window and is intended as a locally deployable alternative to OpenAI's closed API models (model profile).
Capabilities & benchmarks
gpt-oss-120b is categorized as a reasoning model (model profile). Its architecture consists of a 117B-parameter Mixture-of-Experts design optimized for agentic tasks (model profile). While specific internal benchmark scores are not detailed in the model profile, a Signal + Noise wire item from March 2026 reports that Alibaba's [[Qwen3.5]] 9B