#8 gpt-oss-120b
Over a year old and behind the Chinese MoEs on benchmarks, but the most plug-and-play 128 GB model with a clean license.
- Version
- gpt-oss-120b
- Maker
- OpenAI
- Release date
- 2025-08-04 (Hugging Face repo creation date unless noted)
- Architecture
- MoE, MXFP4 post-trained
- Total params
- 117B
- Active params
- 5.1B
- Context length
- no published figure yet unverified
- License
- Apache-2.0
- Commercial notes
- Fully commercial
- Variants
- gpt-oss-120b
- gpt-oss-20b
- gpt-oss-safeguard-120b/20b
- Memory
- BF16/native: no published figure yet
FP8: no published figure yet
Q8: no published figure yet
Q4: Fits a single 80 GB GPU in MXFP4 (model card) - Hardware
- Single Spark (128GB): Yes
Dual Spark (256GB): Yes
Quad Spark (~512GB): Yes
24GB GPU: No
128GB Mac / Strix: Yes (fits in 80 GB) - Tokens/sec
- no published figure yetSpeeds come from the cited sources. Entries marked "hosted, not local" are cloud API speeds, and "ESTIMATE" entries are estimates. Neither is a local measurement.
- Document Q&A strengths
- Reliable tool calling / reasoning effort control
- Mature runtime support everywhere
- Embedding / reranker pairing
- no published figure yet
- Runtimes
- Ollama: Yes (gpt-oss:120b, model card)
llama.cpp: Yes unverified
vLLM: Yes (model card)
LM Studio: Yes (model card) - Hugging Face
- https://huggingface.co/openai/gpt-oss-120b
DGX Spark configs it fits
- Single Spark (128GB): Yes
- Dual Spark (256GB): Yes
- Quad Spark (~512GB): Yes
Spark fit is estimated from published memory sizes, not tested on Spark hardware.
Sources
- https://huggingface.co/openai/gpt-oss-120b
- https://llmcheck.net/blog/state-of-open-source-local-llms-october-2026/
Last updated Oct 24, 2026. Fields tagged unverified lack a primary source. "No published figure yet" means we found no source, and we don't guess. Spark fit is estimated from published memory sizes, not tested on Spark hardware. Ollama listings may be cloud-only tags; check this for Kimi K3 and DeepSeek V4.1 before pulling.
The RAGShelf weekly
New local RAG tools, releases, and tested setups. One email a week, no spam.