furiosa-ai/gpt-oss-20b

TEXT GENERATIONConcurrent Unit Cost:1Model Size:20BQuant:FP8Context Size:32kPublished:Jun 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

furiosa-ai/gpt-oss-20b is a 20 billion parameter Mixture-of-Experts (MoE) transformer model, based on OpenAI's gpt-oss-20b, optimized for lower-latency and on-device reasoning tasks. It features native support for tool calling and configurable reasoning effort, emitting responses in the harmony format which separates chain-of-thought from the final answer. This model is specifically packaged with a Furiosa Executable Bundle (FXB) for efficient deployment on FuriosaAI RNGD hardware.

Loading preview...

Overview of furiosa-ai/gpt-oss-20b

This model is a 20 billion parameter Mixture-of-Experts (MoE) transformer, derived from OpenAI's gpt-oss-20b, designed for efficient reasoning in lower-latency and on-device environments. It is distributed with a Furiosa Executable Bundle (FXB) for optimized execution on FuriosaAI RNGD hardware using Furiosa-LLM.

Key Capabilities

  • Advanced Reasoning: The model first generates a chain of thought, then a final answer, using the harmony response format. Reasoning depth is controllable via reasoning_effort (low, medium, high).
  • Tool Calling: Supports function calling through the openai tool-call parser, enabling interaction with external tools.
  • Quantization: MoE expert weights are quantized to MXFP4 for efficiency, while other components maintain higher precision.
  • Hardware Optimization: Designed to run on a single FuriosaAI RNGD card, utilizing a tensor-parallel size of 8 PEs.

Good for

  • Applications requiring on-device or low-latency reasoning capabilities.
  • Scenarios where explicit chain-of-thought is beneficial for transparency and debugging.
  • Integrating tool use and function calling into AI workflows.
  • Deployment on FuriosaAI RNGD hardware for accelerated inference.