GPT-OSS 120B quants compared
17 GGUF builds by real file size, probed from bartowski/openai_gpt-oss-120b-GGUF on Hugging Face (2026-09-02). 117B params, 5.1B active.
Download GPT-OSS 120B Q4_K_M (58.53 GB across 2 shards) — it fits 96 GB of memory with 16k context. With Ollama: ollama run gpt-oss:120b
Every GPT-OSS 120B quant by real file size
| Quant | Weights | + KV (16k) | Total | Fits comfortably in | Quality |
|---|---|---|---|---|---|
| BF16 | 60.88 GB | 6.0 GB | 66.9 GB | 96 GB | Full precision (lossless) |
| Q8_0 | 59.03 GB | 6.0 GB | 65.0 GB | 96 GB | Near-lossless |
| Q6_K | 58.94 GB | 6.0 GB | 64.9 GB | 96 GB | Excellent |
| Q4_K_M * | 58.53 GB | 6.0 GB | 64.5 GB | 96 GB | High — the default pick |
| Q4_K_L | 58.73 GB | 6.0 GB | 64.7 GB | 96 GB | High |
| Q4_1 | 58.43 GB | 6.0 GB | 64.4 GB | 96 GB | High |
| IQ4_NL | 58.4 GB | 6.0 GB | 64.4 GB | 96 GB | High |
| IQ4_XS | 58.4 GB | 6.0 GB | 64.4 GB | 96 GB | High |
| Q4_0 | 58.34 GB | 6.0 GB | 64.3 GB | 96 GB | High |
| Q3_K_XL | 58.57 GB | 6.0 GB | 64.6 GB | 96 GB | Acceptable — visible loss |
| IQ3_M | 58.4 GB | 6.0 GB | 64.4 GB | 96 GB | Acceptable — visible loss |
| Q3_K_M | 58.4 GB | 6.0 GB | 64.4 GB | 96 GB | Acceptable — visible loss |
| Q3_K_S | 58.39 GB | 6.0 GB | 64.4 GB | 96 GB | Acceptable — visible loss |
| Q3_K_L | 58.3 GB | 6.0 GB | 64.3 GB | 96 GB | Acceptable — visible loss |
| Q2_K_L | 58.67 GB | 6.0 GB | 64.7 GB | 96 GB | Experimental — not ranked — never recommended |
| Q2_K | 58.4 GB | 6.0 GB | 64.4 GB | 96 GB | Experimental — not ranked — never recommended |
| IQ2_M | 58.38 GB | 6.0 GB | 64.4 GB | 96 GB | Experimental — not ranked — never recommended |
* default pick. Weights = real GGUF file sizes from bartowski/openai_gpt-oss-120b-GGUF (probed 2026-09-02). KV = fp16 estimate; a q8_0 cache roughly halves it. "Comfortable" = weights + KV within 90% of memory.
Best GPT-OSS 120B quant by memory
| Memory | Recommended quant | Total (16k ctx) |
|---|---|---|
| 96 GB | BF16 | 66.9 GB |
Why we don't rank GPT-OSS 120B's 2-bit quants
Quants at 2 bits per weight or below (Q2_K, IQ2, IQ1, TQ1) cut file size by roughly half versus Q4, but the quality collapse is steep and non-linear: perplexity spikes, instruction-following degrades, and hallucinations rise. A model that answers faster but wrong is not a smaller model — it is a worse one. ModelFit lists these builds for completeness but never ranks or recommends them.
Frequently asked questions
What is the best quantization of GPT-OSS 120B?
Q4_K_M is the default pick: 58.53 GB of weights, high — the default pick quality, fitting comfortably in 96 GB of memory (weights + 16k context KV-cache). Go Q6_K or Q8_0 if you have headroom.
How much memory does GPT-OSS 120B need?
At Q4_K_M, GPT-OSS 120B needs 58.53 GB for the weights plus ~6.0 GB of KV-cache at 16k context — about 64.5 GB total, so a 96 GB card or Mac (90% usable budget) runs it comfortably.
Should I use a Q2_K or IQ2 quant of GPT-OSS 120B?
No. GPT-OSS 120B at 2 bits per weight is a visibly worse model — quality collapse at that bitrate is steep, not gradual. If only a 2-bit build fits your memory, run a smaller model at Q4_K_M instead. ModelFit lists these builds but never recommends them.
Cite this page
ModelFit: GPT-OSS 120B quantization comparison (real GGUF file sizes). https://modelfit.io/quant-compare/gpt-oss-120b/ (data probed 2026-09-02, CC BY 4.0).