GPT-OSS
GPT-OSS is an open-source or open-weight GPT-series language model released by OpenAI, belonging to the large language model (LLM) category. The model supports natural language understanding and generation, and can be applied to tasks such as text dialogue, content creation, and code assistance. It can be integrated into third-party platforms or deployed in local environments by developers, balancing flexibility with customizability.
Core Facts
Timeline (last 90 days)
GPT-OSS is embedded into the Codex CLI via the --oss flag, with the 20B version running on consumer hardware and the 120B version being server-grade.
Frames of Truth's AI pipeline uses Qwen 3.8 on Groq for visual analysis, GPT-OSS on Groq for story generation, Llama 3.3 on OpenRouter as a narrative fallback, and the Gemini REST API as a visual fallback.
In Qwen3.5, GPT-OSS, and OLMo-3.1, removing the "source endorsement" direction using the mean difference method reduced compliance with incorrect sources by 64 to 78 percentage points, while removing the "user endorsement" direction reduced compliance by at most 11 percentage points.
Ollama free user credits can be used for some cloud models, including Gemma 3 (approximately 4.3 billion parameters), GPT-OSS (120 billion parameters), and Nemotron.
In certain Attention and MoE modules of GPT-OSS, AI-generated implementations are 1.5 to 1.8 times faster than versions written by human experts (applicable only to the selected modules).
All Facts (13)
GPT-OSS采用的是开放权重（Open Weights）策略，仅公开模型参数权重文件，不包含训练数据集和完整训练代码
90%UnverifiedGPT-OSS通过Ollama运行默认是完全本地化的，不需要联网
90%UnverifiedGPT-OSS原生支持FP4混合精度，无需额外量化即可大幅压缩显存占用
90%UnverifiedGPT-OSS完整支持工具调用（Function Calling）和结构化输出等企业级功能
85%UnverifiedGPT-OSS采用了与O4模型完全相同的三阶段训练流程：模型预训练、全量指令微调、RL强化学习后训练
80%UnverifiedGPT-OSS采用了比DeepSeek R1更严格的无监督COT（Chain-of-Thought）对齐方法
75%Partially VerifiedGPT-OSS支持通过系统提示词手动设置推理强度，可在高推理强度和低推理强度之间切换
65%Partially VerifiedGPT-OSS adopts a MoE (Mixture of Experts) architecture combined with Grouped Multi-Query Attention (GQA).
65%UnverifiedGPT-OSS is embedded into the Codex CLI via the --oss flag, with the 20B version running on consumer hardware and the 120B version being server-grade.
50%UnverifiedFrames of Truth's AI pipeline uses Qwen 3.8 on Groq for visual analysis, GPT-OSS on Groq for story generation, Llama 3.3 on OpenRouter as a narrative fallback, and the Gemini REST API as a visual fallback.
50%UnverifiedIn Qwen3.5, GPT-OSS, and OLMo-3.1, removing the "source endorsement" direction using the mean difference method reduced compliance with incorrect sources by 64 to 78 percentage points, while removing the "user endorsement" direction reduced compliance by at most 11 percentage points.
50%UnverifiedOllama free user credits can be used for some cloud models, including Gemma 3 (approximately 4.3 billion parameters), GPT-OSS (120 billion parameters), and Nemotron.
50%UnverifiedIn certain Attention and MoE modules of GPT-OSS, AI-generated implementations are 1.5 to 1.8 times faster than versions written by human experts (applicable only to the selected modules).
50%