gpt-oss:20b-cloud

Feature highlights — shared with gpt-oss-120b

• Agentic capabilities — native function calling, web browsing, Python tool calls, and structured outputs.
• Full chain-of-thought — complete access to the model's reasoning process for easier debugging and increased trust.
• Configurable reasoning effort — adjust reasoning effort (low, medium, high) based on the use case and latency needs.
• Fine-tunable — fully customize to your specific use case through parameter fine-tuning.
• Apache 2.0 license — build freely without copyleft restrictions or patent risk.

Quantization

The models are post-trained with MoE weights quantized to MXFP4 format (4.25 bits per parameter). The MoE weights cover 90+% of the total parameter count, so the 20B model runs on systems with as little as 16GB memory. Ollama supports MXFP4 natively without additional quantization.

Best practices

• Low latency — the 20B variant is designed for lower latency and specialized use cases.
• Reasoning effort — tune low/medium/high to balance quality against response time.
• Local or cloud — :20b-cloud serves it on Ollama's infrastructure; the local tag runs on ~16GB systems.