Model benchmarks
gpt-oss-20b-UD-Q4_K_XL local LLM performance
As of September 2026, gpt-oss-20b-UD-Q4_K_XL runs at up to 124.0 tok/s for local inference (best of 3 community benchmark runs across 2 GPUs).
Model size
20B
Peak speed
124.0 tok/s
Average speed
121.4 tok/s
Avg PP
1579.1 tok/s
Min memory
n/a
Max context
131,072 tokens
Avg output / run
11,505 tokens
Avg runtime / run
1m 40s
Avg quality
57.0
Benchmark runs
3
GPUs tested
2
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 3 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 | 43.5 | 57.0 | 68.5 |
| Agent Workflow | 70.5 | 77.2 | 86.2 |
| Code Generation | 3.2 | 33.2 | 64.2 |
| Role Play & Narrative | 59.5 | 63.2 | 66.0 |
| Research & Analysis | 19.3 | 54.6 | 75.6 |
Performance by hardware and tool
Every hardware/tool/quantization combination gpt-oss-20b-UD-Q4_K_XL has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5060 | llama.cpp | Q4_K | 124.0 tok/s | 123.6 tok/s | n/a | 131,072 tokens | 64.7 | 2 |
| NVIDIA GeForce RTX 4070 | llama.cpp | Q4_K | 117.0 tok/s | 117.0 tok/s | n/a | 8,192 tokens | 41.7 | 1 |
Benchmark runs
All 3 gpt-oss-20b-UD-Q4_K_XL runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is gpt-oss-20b-UD-Q4_K_XL good for coding?
- In our benchmarks, gpt-oss-20b-UD-Q4_K_XL scores 33.2/100 for coding. It runs at about 121.4 tok/s, so if you want more speed, Qwen3.6-35B-A3B-Q4_K_XL is faster (~156.9 tok/s) and still scores well for coding (82.2/100).
- Is gpt-oss-20b-UD-Q4_K_XL good for agentic (tool-using) tasks?
- In our benchmarks, gpt-oss-20b-UD-Q4_K_XL scores 77.2/100 for agentic workflows. It runs at about 121.4 tok/s, so if you want more speed, Tiel-Coder-35B-A3B-Q4_K_S is faster (~162.4 tok/s) and still scores well for agentic workflows (87.8/100).
- How fast is gpt-oss-20b-UD-Q4_K_XL for local inference?
- Across 3 community benchmark runs, gpt-oss-20b-UD-Q4_K_XL reaches up to 124.0 tok/s and averages 121.4 tok/s, with the fastest results on NVIDIA GeForce RTX 5060.
- Which tools have been used to run gpt-oss-20b-UD-Q4_K_XL?
- Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.