Best larger-system route
Studio 96
The smallest system in the range that reaches this model's preferred working allowance.
Explore Studio 96OpenAI model
GPT-OSS 120B Hardware Requirements & Compatibility
OpenAI states that GPT-OSS 120B fits on one 80GB GPU. In this range, Studio 96 is the first straightforward single-GPU product; 32GB products are appropriate for GPT-OSS 20B rather than the named 120B checkpoint.
- Model version
openai/gpt-oss-120b- Family and variant
- GPT-OSS 120B · gpt-oss-120b
- Source version
b5c939de8f75- Source updated
- 26 August 2025
OpenAI GPT-OSS 120B
openai/gpt-oss-120b - Minimum GPU memory
- One 80GB GPU
- Recommended hardware
- One 96GB GPU for additional operating headroom
- Licence
- Apache License 2.0
- Useful for
- Language & reasoning, Coding & agents
Buyer verdict
Where GPT-OSS 120B is a sensible fit
GPT-OSS 120B is well suited to tool use, structured outputs, reasoning and fine-tuning where Apache licensing and a one-80GB-GPU deployment are important.
These figures apply to the named model version. Quantisation, fine-tuning, context length, image resolution, batch size and serving software can materially change the hardware needed.
Hardware requirements
- Minimum
- One 80GB GPU
- Recommended
- One 96GB GPU for additional operating headroom
OpenAI documents this as the serving target for GPT-OSS 120B.
Context and concurrency still determine usable service capacity.
Technical specification
GPT-OSS 120B model and hardware facts
Specifications shown for source version
b5c939de8f754692c1647ca79fbf85e8c1e70f8a, updated
26 August 2025.
- Architecture
- MoE, 117B total / 5.1B active
- Context
- 131,072 tokens
- Native format
- MXFP4 MoE weights
- Official GPU floor
- One 80GB GPU
- Response format
- Harmony
- Licence
- Apache 2.0
Product compatibility
GPT-OSS 120B compatibility across all 11 GPU systems
- Systems that fit without model splitting
- 7
- Systems needing multi-GPU validation
- 0
A system is listed as fitting when its GPU memory meets the requirement shown above. Speed, usable context, batch size and concurrent users still need testing with the final model and software configuration.
sme workstation
Team 32
- Per GPU
- 32GB
- Total GPU memory
- 32GB
- GPU count
- 1
This system provides 32GB per GPU and 32GB in total. The model requires 80GB on one GPU.
sme workstation
Company 64
- Per GPU
- 32GB
- Total GPU memory
- 64GB
- GPU count
- 2
This system provides 32GB per GPU and 64GB in total. The model requires 80GB on one GPU.
sme workstation
Studio 96
- Per GPU
- 96GB
- Total GPU memory
- 96GB
- GPU count
- 1
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the model developer's published requirement. Confirm speed, context, batch size and concurrent users before purchase.
sme workstation
Studio 192
- Per GPU
- 96GB
- Total GPU memory
- 192GB
- GPU count
- 2
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the model developer's published requirement. Confirm speed, context, batch size and concurrent users before purchase.
pcie rack
Value Rack 128
- Per GPU
- 32GB
- Total GPU memory
- 128GB
- GPU count
- 4
This system provides 32GB per GPU and 128GB in total. The model requires 80GB on one GPU.
pcie rack
Value Rack 256
- Per GPU
- 32GB
- Total GPU memory
- 256GB
- GPU count
- 8
This system provides 32GB per GPU and 256GB in total. The model requires 80GB on one GPU.
pcie rack
Enterprise 384
- Per GPU
- 96GB
- Total GPU memory
- 384GB
- GPU count
- 4
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the model developer's published requirement. Confirm speed, context, batch size and concurrent users before purchase.
pcie rack
Enterprise 768
- Per GPU
- 96GB
- Total GPU memory
- 768GB
- GPU count
- 8
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the model developer's published requirement. Confirm speed, context, batch size and concurrent users before purchase.
pcie rack
H200 1.1TB
- Per GPU
- 141GB
- Total GPU memory
- 1,128GB
- GPU count
- 8
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the model developer's published requirement. Confirm speed, context, batch size and concurrent users before purchase.
frontier partner
Frontier Native 2.3TB
- Per GPU
- 288GB
- Total GPU memory
- 2,304GB
- GPU count
- 8
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the model developer's published requirement. Confirm speed, context, batch size and concurrent users before purchase.
frontier partner
Frontier Rack 20TB
- Per GPU
- 288GB
- Total GPU memory
- 20,736GB
- GPU count
- 72
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the model developer's published requirement. Confirm speed, context, batch size and concurrent users before purchase.
Deployment reality
Strengths, limits and runtime route
What it is good at
- Official single-80GB-GPU deployment statement.
- Configurable reasoning effort and agentic tool-use support.
- Apache 2.0 licence and official reference implementations.
- OpenAI-compatible serving routes through vLLM and other supported tools.
Where to be cautious
- The Harmony response format is required for correct operation.
- Reasoning traces are not intended to be shown directly to end users.
- An 80GB fit statement does not include every context, concurrency or fine-tuning case.
- Repository files contain several representations, so summing the entire repository overstates the serving artefact.
Serving software
- vLLM
- Transformers
- Ollama
- llama.cpp
- OpenAI reference implementation
Before installation
- Use the official Harmony format and a runtime release that explicitly supports GPT-OSS.
- Treat the 80GB statement as a minimum for serving and reserve space for the real context and concurrency target.
- Use GPT-OSS 20B for the 32GB product tier when its quality is sufficient.
- Fine-tuning has a different memory profile from inference and needs its own plan.
System requirements
- GPU layout
- The minimum can fit on one GPU; multiple GPUs may still be useful for replicas or throughput.
- Context and cache
- The model context ceiling is not a guaranteed serving target. KV cache, batch size and concurrent sessions need separate capacity tests.
- System RAM
- Size system memory for model loading, runtime overhead, preprocessing and any CPU offload used by the final configuration.
- Storage
- Allow space for the pinned checkpoint, runtime images, caches, logs and at least one rollback version.
- Serving software
- Validate the exact checkpoint with vLLM, Transformers, Ollama, llama.cpp, OpenAI reference implementation before acceptance.
- Representative workload
- Benchmark representative prompts or media at the required context, quality, latency and concurrency.
Performance: speed, usable context and concurrency depend on the selected system, software and workload. No benchmark is quoted on this page.
Commercial and legal boundary
Apache License 2.0
Commercial use: permitted
Apache 2.0 permits commercial use subject to notices and the intended application's obligations.
Always retain the applicable notices and recheck the live terms for the intended organisation, territory, use and distribution route. Obtain legal advice where required; the official licence governs use.
Read the official licenceTechnical questions
GPT-OSS 120B deployment FAQ
Which product can run GPT-OSS 120B on one GPU?
Studio 96 is the first product in the range with more than the official 80GB GPU floor.
Can Team 32 run GPT-OSS?
Team 32 is better matched to GPT-OSS 20B. A third-party 120B conversion is a different model version and should not be described as the official 120B release.
Can GPT-OSS 120B be fine-tuned on the same server?
OpenAI documents fine-tuning routes, but training memory differs from inference. The method, optimiser, sequence length and adapter or full-weight scope must be sized separately.
What a complete GPT-OSS 120B deployment needs
GPU memory is only one part of the system. Storage, data access, serving software, monitoring and administrator handover also affect a reliable deployment.
