Model benchmarks

GRM-3.2-Sky-ONYX-quality local LLM performance

As of October 2026, GRM-3.2-Sky-ONYX-quality runs at up to 12.3 tok/s for local inference (best of 15 community benchmark runs across 1 GPU).

ik_llama.cpp:cpullama.cpp
ShareRedditX

Model size

Unknown

Peak speed

12.3 tok/s

Average speed

11.3 tok/s

Avg PP

53.2 tok/s

Min memory

n/a

Max context

262,144 tokens

Avg output / run

15,665 tokens

Avg runtime / run

26m 46s

Avg quality

78.0

Benchmark runs

15

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 15 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)
Overall68.978.082.6
Agent Workflow65.976.687.7
Code Generation64.773.378.9
Role Play & Narrative74.985.892.5
Research & Analysis52.376.384.8

Performance by hardware and tool

Every hardware/tool/quantization combination GRM-3.2-Sky-ONYX-quality has been benchmarked on, ranked by peak token generation speed. Last updated October 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
Intel Graphicsik_llama.cpp:cpu—12.3 tok/s11.2 tok/sn/a262,144 tokens79.514
Intel Graphicsllama.cpp—12.1 tok/s12.1 tok/sn/a262,144 tokens56.41

Benchmark runs

All 15 GRM-3.2-Sky-ONYX-quality runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Frequently asked questions

Is GRM-3.2-Sky-ONYX-quality good for coding?
In our benchmarks, GRM-3.2-Sky-ONYX-quality scores 73.3/100 for coding. It runs at about 11.3 tok/s, so if you want more speed, Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-NM-DAU-NEO-MAX-MTP-GGUF-Q6_K is faster (~34.2 tok/s) and still scores well for coding (85.9/100).
Is GRM-3.2-Sky-ONYX-quality good for agentic (tool-using) tasks?
In our benchmarks, GRM-3.2-Sky-ONYX-quality scores 76.6/100 for agentic workflows. It runs at about 11.3 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 GRM-3.2-Sky-ONYX-quality for local inference?
Across 15 community benchmark runs, GRM-3.2-Sky-ONYX-quality reaches up to 12.3 tok/s and averages 11.3 tok/s, with the fastest results on Intel Graphics.
Which tools have been used to run GRM-3.2-Sky-ONYX-quality?
Benchmarks were submitted using ik_llama.cpp:cpu, llama.cpp. Results are community-contributed and updated as new runs arrive.