Model benchmarks

ptq1 local LLM performance

As of September 2026, ptq1 runs at up to 49.2 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).

llama.cpp
ShareRedditX

Model size

Unknown

Peak speed

49.2 tok/s

Average speed

49.1 tok/s

Avg PP

401.6 tok/s

Min memory

3.4 GB

Max context

8.192 tokens

Avg output / run

15.367 tokens

Avg runtime / run

5m 25s

Avg quality

61.2

Benchmark runs

2

GPUs tested

1

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 2 runs. The low and high columns show how much the judge’s score varies between runs, and need at least two runs to display.

TaskP5 (low)AvgP95 (high)
Overall60.161.262.4
Agent Workflow70.173.176.1
Code Generation6.612.117.5
Role Play & Narrative86.086.086.0
Research & Analysis71.673.876.1

Performance by hardware and tool

Every hardware/tool/quantization combination ptq1 has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
NVIDIA GeForce RTX 4070llama.cpp49.2 tok/s49.1 tok/s3.4 GB8.192 tokens61.22

Benchmark runs

All 2 ptq1 runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Frequently asked questions

Is ptq1 good for coding?
In our benchmarks, ptq1 scores 12.1/100 for coding. It runs at about 49.1 tok/s, so if you want more speed, unsloth/Qwen3.8-27B-GGUF:IQ3_S is faster (~104.3 tok/s) and still scores well for coding (83.0/100).
Is ptq1 good for agentic (tool-using) tasks?
In our benchmarks, ptq1 scores 73.1/100 for agentic workflows. It runs at about 49.1 tok/s, so if you want more speed, granite-4.2-8b-Q4_K_L is faster (~71.6 tok/s) and still scores well for agentic workflows (92.9/100).
How fast is ptq1 for local inference?
Across 2 community benchmark runs, ptq1 reaches up to 49.2 tok/s and averages 49.1 tok/s, with the fastest results on NVIDIA GeForce RTX 4070.
How much memory does ptq1 need?
The leanest observed configuration used about 3.4 GB of memory.
Which tools have been used to run ptq1?
Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.