Model record
gpt-oss-120b
OpenAI · Apache 2.0 · verified
What this model is
OpenAI reports 117B total parameters and 5.1B active parameters per token.
The architecture is mixture of experts. Memory must hold the full weight set, while active parameters affect the speed heuristic. Experts can be offloaded to system memory, but that trades speed for capacity.
Memory and quantization
Decimal GB. The MXFP4 GGUF file is 63.39 GB; the checked GGUF repository does not publish a separate Q8 build.
| Quantization | Published files | 8K estimate | Smallest comfortable GPU | Smallest comfortable Mac | Context at smallest GPU / Mac | Speed estimate | Run commands |
|---|---|---|---|---|---|---|---|
| MXFP4 | 63.39 GB | 65.19 GB | NVIDIA RTX PRO 6000 Blackwell 96GB · 96 GB VRAM | Apple Mac Studio M3 Ultra 96GB · 96 GB unified memory | 131,072 / 131,072 tokens | Estimate · Moderate relative class | ollama run gpt-oss:120b lms get "gpt-oss-120b"@mxfp4 |
Estimate formula: published weight files + 0.3067 GB per 8K tokens × context / 8K + 1.5 GB runtime overhead. Sizes are decimal GB, not GiB. At 8K context, the smallest listed build is about 65.2 GB before OS and application reserve.
Conservative FP16 estimate from the public config. Alternating full-attention layers use 8 KV heads; local layers are capped by a 128-token sliding window.
Ollama tags were checked against the linked Ollama library pages on 2026-11-02. LM Studio's lms get suffix selects a file when the search results expose that exact quantization; confirm the local key with lms ls.
KV method source: https://huggingface.co/openai/gpt-oss-120b/raw/main/config.json
Speed heuristic
Not benchmarked here. The calculator uses 5.1B active parameters only as a relative speed heuristic.
Active path: 5.1B parameters. This is a relative speed signal only; it is not a tokens/sec benchmark.
If the full weights do not fit in accelerator memory but do fit in system memory, the calculator marks the result as a tight CPU-offload fit. Expect substantially slower generation and verify the runtime's MoE offload support.
Runtimes
- Ollama
- llama.cpp
- vLLM
ollama run gpt-oss:120blms get "gpt-oss-120b"@mxfp4
lms load <model-key-from-lms-ls> --gpu max --context-length 8192License restrictions
Apache 2.0 permits commercial use and modification; retain notices and comply with the patent and termination terms.
This is a plain-language summary, not legal advice. Read the linked license before commercial deployment or redistribution.
Known issues
- OpenAI positions this model for a single 80 GB accelerator; 64 GB systems are below the MXFP4 weight requirement before KV cache.
- The Harmony response format is required.
If it does not fit, try these smaller records
These alternatives have a lower recommended memory target in the current catalog. Open each page to compare the quantization-specific fit before downloading.
Qwen3.6 35B-A3B
MoE · starting at 25 GB
Gemma 4 31B
Dense · starting at 23 GB
Qwen3.8 27B
Dense · starting at 20 GB
gpt-oss-20b
MoE · starting at 14 GB
FAQ
Model-specific questions
What is the smallest local file for gpt-oss-120b?
The smallest published weight package in this record is MXFP4 at 63.39 GB. The 8K runtime estimate adds 0.31 GB of KV cache and 1.5 GB overhead, for about 65.2 GB before operating-system reserve.
How much memory do the published quantizations need at 8K?
MXFP4 about 65.2 GB at 8K. These are calculated estimates using the published weight sizes and the model's KV-cache assumption, not live memory measurements.
Which GPU and Mac configurations are comfortable for gpt-oss-120b?
Against the current catalog, the smallest configurations that clear the 20% headroom target are: MXFP4: NVIDIA RTX PRO 6000 Blackwell 96GB. A larger configuration may still be preferable for longer context or faster memory bandwidth.
What is the right Ollama or LM Studio command for gpt-oss-120b?
Use the command attached to the specific quantization in the table. Ollama tags are taken from the checked Ollama library pages. For LM Studio, use lms get with the displayed quantization suffix, then load the resulting local key from lms ls with --gpu max and the desired context length.
What should I run instead if gpt-oss-120b does not fit?
Smaller catalog alternatives include Qwen3.6 35B-A3B, Gemma 4 31B, Qwen3.8 27B, gpt-oss-20b. Compare their quantization tables and minimum memory before choosing; a smaller parameter count does not always mean a proportionally smaller download.
Sources
- https://huggingface.co/openai/gpt-oss-120b
- https://huggingface.co/ggml-org/gpt-oss-120b-GGUF
- https://ollama.com/library/gpt-oss
- https://huggingface.co/openai/gpt-oss-120b/raw/main/config.json
This page is an independent compatibility guide, not official documentation. Verify the primary source before downloading or deploying a model.