Is gpt-oss-20b-UD-Q8_K_XL good for agentic (tool-using) tasks?

In our benchmarks, gpt-oss-20b-UD-Q8_K_XL scores 80.1/100 for agentic workflows. It runs at about 129.4 tok/s, so if you want more speed, Tiel-Coder-35B-A3B-MLX-oQ4e-MTP is faster (~151.8 tok/s) and still scores well for agentic workflows (83.0/100).

How fast is gpt-oss-20b-UD-Q8_K_XL for local inference?

Across 5 community benchmark runs, gpt-oss-20b-UD-Q8_K_XL reaches up to 176.2 tok/s and averages 129.4 tok/s, with the fastest results on NVIDIA GeForce RTX 5060.

How much memory does gpt-oss-20b-UD-Q8_K_XL need?

The leanest observed configuration used about 16.4 GB of memory (quantizations tested: Q8_K).

Which tools have been used to run gpt-oss-20b-UD-Q8_K_XL?

Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.

Model benchmarks

gpt-oss-20b-UD-Q8_K_XL local LLM performance

As of October 2026, gpt-oss-20b-UD-Q8_K_XL runs at up to 176.2 tok/s for local inference (best of 5 community benchmark runs across 1 GPU).

llama.cppQ8_K

ShareReddit X

Model size

20B

Peak speed

176.2 tok/s

Average speed

129.4 tok/s

Avg PP

1434.4 tok/s

Min memory

16.4 GB

Max context

131,072 tokens

Avg output / run

16,078 tokens

Avg runtime / run

Avg quality

73.4

Benchmark runs

GPUs tested

Quality by task

Average LLM-judged quality (0–100) with the run-to-run spread shown as a P5–P95 band, overall and for each benchmark task, across all 5 runs. The low and high columns show how much the judge’s score varies between runs, and need at least two runs to display.

Task	P5 (low)	Avg	P95 (high)
Overall	68.9	73.4	77.0
Agent Workflow	72.1	80.1	87.0
Code Generation	67.3	71.0	74.9
Role Play & Narrative	55.8	67.2	74.6
Research & Analysis	69.4	75.1	80.7

Performance by hardware and tool

Every hardware/tool/quantization combination gpt-oss-20b-UD-Q8_K_XL has been benchmarked on, ranked by peak token generation speed. Last updated October 2026.

Hardware	Tool	Quant	Best tok/s	Avg tok/s	Memory	Context	Quality	Runs
NVIDIA GeForce RTX 5060	llama.cpp	Q8_K	176.2 tok/s	129.4 tok/s	16.4 GB	131,072 tokens	73.4	5

Benchmark runs

All 5 gpt-oss-20b-UD-Q8_K_XL runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Open run page →

Frequently asked questions

Is gpt-oss-20b-UD-Q8_K_XL good for agentic (tool-using) tasks?: In our benchmarks, gpt-oss-20b-UD-Q8_K_XL scores 80.1/100 for agentic workflows. It runs at about 129.4 tok/s, so if you want more speed, Tiel-Coder-35B-A3B-MLX-oQ4e-MTP is faster (~151.8 tok/s) and still scores well for agentic workflows (83.0/100).
How fast is gpt-oss-20b-UD-Q8_K_XL for local inference?: Across 5 community benchmark runs, gpt-oss-20b-UD-Q8_K_XL reaches up to 176.2 tok/s and averages 129.4 tok/s, with the fastest results on NVIDIA GeForce RTX 5060.
How much memory does gpt-oss-20b-UD-Q8_K_XL need?: The leanest observed configuration used about 16.4 GB of memory (quantizations tested: Q8_K).
Which tools have been used to run gpt-oss-20b-UD-Q8_K_XL?: Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.