1× RTX 5090
367tok/sPeak
1 request, synthetic prompt Source
Intelligence9with thinking
Local AI · Recipe · 1× RTX PRO 6000
gpt-oss 21B-A3.6B MXFP4 with llama.cpp on 1× RTX PRO 6000
gpt-oss 21B-A3.6B MXFP4 with llama.cpp on 1× RTX PRO 6000: 287 tok/s according to github.com (dataset as of Nov 3, 2026).
by ggml-org
Engine
- Engine
- llama.cpp lt. Guide #15396
- Quantization
- MXFP4 (GGUF)
- Model family
- gpt-oss
- Context
- —
- Parameters
- 21B MoE / 3.6B active
- Creator
- ggml-org
- GitHub stars
- 133,144
- Repo updated
- Nov 2, 2026
Measurements
Every sourced value of this recipe, each with its condition and source. Bars relative to the largest value in the group.
Decode · 1 request
Prefill by context
What you need
- Hardware
- 1 × NVIDIA RTX PRO 6000 Blackwell
- Weights
- Main weightsggml-org/gpt-oss-20b-GGUF Apache-2.0
- Engine
- llama.cpp lt. Guide #15396
Related recipes
1× RTX 5090
283tok/sPeak
1 request, synthetic prompt Source
Intelligence9with thinking
1× RTX PRO 6000
196tok/sPeak
1 request, synthetic prompt Source
Intelligence12with thinking
1× RTX PRO 6000
171tok/sPeak
1 request, synthetic prompt Source
Intelligence12with thinking
1× RTX 3090
162tok/sPeak
1 request, synthetic prompt Source
Intelligence9with thinking
1× RTX 5090
46.5tok/sPeak
1 request, synthetic prompt Source
Intelligence12with thinking
- Partly in system RAM
Local AI in your company?
In a workshop we work out which models and which hardware fit your tasks, and build the first agent on your infrastructure.