Model benchmarks
ggml-org/gpt-oss-20b-GGUF local LLM performance
As of October 2026, ggml-org/gpt-oss-20b-GGUF runs at up to 144.0 tok/s for local inference (best of 6 community benchmark runs across 2 GPUs).
Model size
20B
Peak speed
144.0 tok/s
Average speed
81.9 tok/s
Avg PP
607.9 tok/s
Min memory
12.1 GB
Max context
32,768 tokens
Avg output / run
22,313 tokens
Avg runtime / run
10m 6s
Avg quality
60.4
Benchmark runs
6
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 6 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 | 36.1 | 60.4 | 73.2 |
| Agent Workflow | 17.4 | 69.5 | 90.8 |
| Code Generation | 0.0 | 37.2 | 78.0 |
| Role Play & Narrative | 52.4 | 62.0 | 71.7 |
| Research & Analysis | 41.2 | 72.9 | 88.1 |
Performance by hardware and tool
Every hardware/tool/quantization combination ggml-org/gpt-oss-20b-GGUF 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 |
|---|---|---|---|---|---|---|---|---|
| AMD Radeon RX 9070/9070 XT/9070 GRE | llama.cpp | — | 144.0 tok/s | 142.1 tok/s | 12.1 GB | 32,768 tokens | 71.4 | 3 |
| Intel(R) Arc(TM) 140V GPU (16GB) | llama.cpp | — | 23.0 tok/s | 21.7 tok/s | n/a | 16,384 tokens | 49.4 | 3 |
Benchmark runs
All 6 ggml-org/gpt-oss-20b-GGUF runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is unsloth/gpt-oss-20b-GGUF good for coding?
- In our benchmarks, unsloth/gpt-oss-20b-GGUF scores 37.2/100 for coding. It runs at about 81.9 tok/s, so if you want more speed, Ornith-1.5-35B-A3B.APEX-I-NanoPlus is faster (~110.5 tok/s) and still scores well for coding (81.1/100).
- Is unsloth/gpt-oss-20b-GGUF good for agentic (tool-using) tasks?
- In our benchmarks, unsloth/gpt-oss-20b-GGUF scores 69.5/100 for agentic workflows. It runs at about 81.9 tok/s, so if you want more speed, qwen3.8-flash-next-iq3_s is faster (~98.1 tok/s) and still scores well for agentic workflows (90.6/100).
- How fast is ggml-org/gpt-oss-20b-GGUF for local inference?
- Across 6 community benchmark runs, ggml-org/gpt-oss-20b-GGUF reaches up to 144.0 tok/s and averages 81.9 tok/s, with the fastest results on AMD Radeon RX 9070/9070 XT/9070 GRE.
- How much memory does ggml-org/gpt-oss-20b-GGUF need?
- The leanest observed configuration used about 12.1 GB of memory.
- Which tools have been used to run ggml-org/gpt-oss-20b-GGUF?
- Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.