# gpt-oss-20b datasheet

> A mixture-of-experts model shipped natively in MXFP4. Only 3.6B parameters are read per token, so it is fast on almost anything with 16 GB.

Canonical: https://thelocalbench.com/models/gpt-oss-20b · Data as of 2026-10-12 · The Local Bench (thelocalbench.com), data as of 2026-10-12, CC BY 4.0

Flags: [verified] checked against primary source · [sourced] cited · [estimate] formula · [sample] placeholder

## Specifications

| Field | Value | Flag | Source |
|---|---|---|---|
| params_total | 20.91 B | verified | https://huggingface.co/api/models/openai/gpt-oss-20b |
| params_active | 3.6 B | sourced | https://huggingface.co/openai/gpt-oss-20b |
| layers | 24 | verified | https://huggingface.co/openai/gpt-oss-20b/blob/main/config.json |
| kv_heads | 8 | verified | https://huggingface.co/openai/gpt-oss-20b/blob/main/config.json |
| q_heads | 64 | verified | https://huggingface.co/openai/gpt-oss-20b/blob/main/config.json |
| head_dim | 64 | verified | https://huggingface.co/openai/gpt-oss-20b/blob/main/config.json |
| hidden_size | 2880 | verified | https://huggingface.co/openai/gpt-oss-20b/blob/main/config.json |
| context | 131072 tokens | verified | https://huggingface.co/openai/gpt-oss-20b/blob/main/config.json |
| sliding_window | 128 tokens | verified | https://huggingface.co/openai/gpt-oss-20b/blob/main/config.json |
| sliding_layers | 12 | verified | https://huggingface.co/openai/gpt-oss-20b/blob/main/config.json |
| experts | 32 experts, 4 active per token | verified | https://huggingface.co/openai/gpt-oss-20b/blob/main/config.json |
| license | Apache-2.0 | verified | https://www.apache.org/licenses/LICENSE-2.0 |

## Memory and quants (total = weights + F16 KV + overhead)

| Quant | bpw | File GB | At 8k | At 32k | At 128k | KLD | 24 GB card, 8k |
|---|---|---|---|---|---|---|---|
| MXFP4 | 4.25 | 12.11 [verified] | 12.9 | 13.7 | 16.8 | — | fit |

## Speed (default quant, 8k context, single stream)

| Device | Verdict | Decode tok/s | Flag |
|---|---|---|---|
| [GeForce RTX 5060 Ti 16 GB](/hardware/rtx-5060-ti-16g.md) | fit | 117.4 | estimate |
| [GeForce RTX 4060 Ti 16 GB](/hardware/rtx-4060-ti-16g.md) | fit | 75.5 | estimate |
| [Radeon RX 7900 XTX 24 GB](/hardware/rx-7900-xtx-24g.md) | fit | 251.6 | estimate |
| [GeForce RTX 3090 24 GB (used)](/hardware/rtx-3090-24g-used.md) | fit | 245.3 | estimate |
| [CPU only · Ryzen 9 9950X + 128 GB DDR5](/hardware/cpu-ryzen-9-9950x-128g.md) | fit | 19.6 | estimate |
| [2× RTX 3090 (used) 48 GB](/hardware/2x-rtx-3090-48g-used.md) | fit | 245.3 | estimate |
| [Mac Studio M5 Max (base)](/hardware/mac-studio-m5-max.md) | fit | 90.4 | estimate |
| [Ryzen AI Max+ 395 mini PC 128 GB](/hardware/ryzen-ai-max-395-128g.md) | fit | 50.3 | estimate |
| [GeForce RTX 4090 24 GB](/hardware/rtx-4090-24g.md) | fit | 264.2 | estimate |
| [NVIDIA DGX Spark FE 128 GB](/hardware/dgx-spark-128g.md) | fit | 75.2 | sourced |
| [MacBook Pro M4 Max 40c · 128 GB](/hardware/macbook-pro-m4-max-128g.md) | fit | 107.3 | estimate |
| [Mac Studio M5 Ultra (base)](/hardware/mac-studio-m5-ultra.md) | fit | 235.9 | estimate |
| [GeForce RTX 5090 32 GB](/hardware/rtx-5090-32g.md) | fit | 469.6 | estimate |
| [Mac Studio M5 Ultra · 512 GB](/hardware/mac-studio-m5-ultra-512g.md) | fit | 235.9 | estimate |

## Quality

LMArena text score: 1287.4 [sourced] (LMArena text, overall, entry 'gpt-oss-20b': 1287.4 (95% CI 1281.0–1293.7, 10,418 votes), leaderboard published 2026-10-08. CC-BY-4.0.)

JSON: https://thelocalbench.com/api/v1/models/gpt-oss-20b.json · Configurator: https://thelocalbench.com/configurator?model=gpt-oss-20b