MODEL / OPENAI

gpt-oss-120b

OpenAI targets one 80 GB accelerator. Correct serving requires Harmony format support.

Primärquelle

Gesamtparameter117B5.1B active / token

Geschätzter Speicher~80 GBMXFP4

Kontext128KText

LicenseApache-2.0Primarily English

01 / GEEIGNET FÜR

reasoning
agents
datacenter serving

02 / GESCHÄTZTE PASSFORM

Framework Desktop · Ryzen AI Max+ 395 · 128 GBBequem NVIDIA DGX Spark · 128 GBBequem Mac Studio M3 Ultra · 256 GBBequem

03 / STARTEN MIT

These are runtime examples, not validated installation recipes. Fit estimates assume quantized files; official repositories may hold full-precision weights. MLX, GGUF and TensorRT require matching conversions or engines.

vLLMAdvancedvllm serve openai/gpt-oss-120b --dtype auto

TensorRT-LLMAdvancedtrtllm-serve openai/gpt-oss-120b

llama.cppMediumllama-server -m /path/to/verified-model.gguf --port 9931

SGLangAdvancedpython3 -m sglang.launch_server --model-path openai/gpt-oss-120b --port 30000

Achtung

OpenAI targets one 80 GB accelerator. Correct serving requires Harmony format support.