Quick Facts
| Developer | OpenAI |
| License | Apache 2.0 -- permits commercial use subject to standard Apache terms |
| Sizes | 20B and 120B mixture-of-experts |
| Context Window | 128K tokens (both sizes) |
| Cloud Variants | gpt-oss:20b-cloud, gpt-oss:120b-cloud |
| Modality | Text only; configurable reasoning effort (low/medium/high); native tool use |
| Local Hardware (if self-hosting) | 20B: ~16 GB memory. 120B: ~80 GB-class GPU |
| Best For | General reasoning, agentic pipelines, function calling, structured output |
Why Choose This Model
gpt-oss is a strong general-purpose reasoning and agent model with visible chain-of-thought output, configurable reasoning effort, and native tool calling -- useful for prototyping agent pipelines, validating structured JSON output, and everyday drafting or Q&A work. The 20B tag covers most everyday use at lower cost; the 120B tag is worth the added spend when reasoning quality clearly matters. Because both sizes use standard Ollama tags, code and prompts developed against the cloud variant port to a self-hosted deployment with minimal changes if you later need to move on-premises.
Terminal.Glass Deployment
Your interface, chat history, user accounts, and RAG index stay on your Terminal.Glass host.
When you submit a prompt -- including any retrieved passages from your RAG index -- to
gpt-oss:20b-cloud or
gpt-oss:120b-cloud, the
request is forwarded to Ollama Cloud, where gpt-oss runs, and the response streams back. Cloud
access requires an Ollama account signed in on the Terminal.Glass machine
(ollama signin). Higher
reasoning effort settings and the 120B tag increase per-request latency and token cost relative
to shorter 20B chats. No local GPU is required to run either cloud tag.
Recommended Uses
- Everyday chat, drafting, and long-form writing assistance
- Structured reasoning, function calling, and JSON-output prototyping
- Lightweight document RAG and internal Q&A
- Agent pipeline prototyping and tool-calling validation
- Literature summarization and exploratory analysis on public or low-sensitivity text
Not appropriate for HIPAA-covered data, PCI-regulated payment details, attorney-client material, proprietary source code, export-controlled documents, or other confidential content -- prompts and RAG-retrieved passages are sent to Ollama Cloud during inference. gpt-oss is text-only with no image upload path; teams needing multimodal or vision input should evaluate another model family. Cloud inference is not strictly reproducible across sessions since backend tags and provider behavior can change. gpt-oss is a general-purpose model, not a validated medical device, and should not be used for clinical or diagnostic decisions.
Hardware Guidance
Both gpt-oss cloud tags run entirely on Ollama Cloud, so the Terminal.Glass host only needs to run Open WebUI, the Ollama client, and your document index -- a small VM or mini-PC is enough. Teams that want to self-host instead should plan for roughly 16 GB of memory for the 20B model or an 80 GB-class GPU for the 120B model, since local weights range from 14–65 GB depending on quantization. Confirm current tag availability at the official Ollama listing.
Honest Guidance
- Strict privacy, regulated, or confidential data: route this to a self-hosted deployment instead -- Terminal Glass's cloud routing sends prompts and RAG-retrieved content to Ollama Cloud and is not equivalent to an air-gapped setup.
- Heavy proprietary code or sensitive internal documents: keep inference on hardware you control rather than routing source or contracts through Ollama Cloud.
- Predictable high volume over many months: a local 20B deployment on owned hardware often becomes more economical than sustained cloud token spend, especially for automated pipelines.
- Fixed model versions or audit-grade reproducibility: a pinned local gpt-oss build is more reliable for research, compliance review, or regression testing than a cloud tag that may change behavior.
- Multimodal or vision workloads: gpt-oss is text-only. Teams needing image or document screenshot analysis should evaluate a multimodal model family instead.
- Clinical or diagnostic use: gpt-oss is a general-purpose model, not a validated medical device -- do not use it for patient care decisions without clinical governance your compliance team approves.
Learn More
Full technical reference, licensing detail, and local deployment options for this model family are documented on NoCloudGPT. Confirm current tags and availability at ollama.com/library/gpt-oss.