Is google/gemma-4-26b-a4b-qat good for coding?

In our benchmarks, google/gemma-4-26b-a4b-qat scores 65.4/100 for coding. It runs at about 66.9 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 google/gemma-4-26b-a4b-qat good for agentic (tool-using) tasks?

In our benchmarks, google/gemma-4-26b-a4b-qat scores 61.9/100 for agentic workflows. It runs at about 66.9 tok/s, so if you want more speed, swift-iq2_xs is faster (~91.5 tok/s) and still scores well for agentic workflows (91.4/100).

How fast is google/gemma-4-26b-a4b-qat for local inference?

Across 8 community benchmark runs, google/gemma-4-26b-a4b-qat reaches up to 157.0 tok/s and averages 66.9 tok/s, with the fastest results on NVIDIA GeForce RTX 4090.

How much memory does google/gemma-4-26b-a4b-qat need?

The leanest observed configuration used about 18.5 GB of memory (quantizations tested: 4bit, Q4_0).

Which tools have been used to run google/gemma-4-26b-a4b-qat?

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

Model benchmarks

google/gemma-4-26b-a4b-qat local LLM performance

As of October 2026, google/gemma-4-26b-a4b-qat runs at up to 157.0 tok/s for local inference (best of 8 community benchmark runs across 4 GPUs).

LM Studio4bitQ4_0

ShareReddit X

Model size

26B

Peak speed

157.0 tok/s

Average speed

66.9 tok/s

Avg PP

1426.0 tok/s

Min memory

18.5 GB

Max context

65,536 tokens

Avg output / run

17,429 tokens

Avg runtime / run

6m 26s

Avg quality

66.4

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 8 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	61.1	66.4	71.7
Agent Workflow	32.5	61.9	78.6
Code Generation	60.7	65.4	71.8
Role Play & Narrative	68.2	77.8	84.5
Research & Analysis	51.6	60.7	70.8

Performance by hardware and tool

Every hardware/tool/quantization combination google/gemma-4-26b-a4b-qat 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 4090	LM Studio	—	157.0 tok/s	157.0 tok/s	n/a	8,192 tokens	61.6	1
Apple M5 Max	LM Studio	4bit	109.4 tok/s	109.4 tok/s	18.5 GB	65,536 tokens	67.5	1
Apple M5 Max	LM Studio	—	108.4 tok/s	108.4 tok/s	n/a	65,536 tokens	62.7	1
NVIDIA GeForce RTX 4070 Ti SUPER	LM Studio	—	51.4 tok/s	51.4 tok/s	n/a	16,384 tokens	68.8	1
Apple M5	LM Studio	—	33.8 tok/s	33.8 tok/s	n/a	8,192 tokens	60.9	1
CPU only	LM Studio	Q4_0	30.3 tok/s	29.9 tok/s	n/a	65,536 tokens	69.1	2
CPU only	LM Studio	—	15.7 tok/s	15.7 tok/s	n/a	32,768 tokens	71.8	1

Benchmark runs

All 8 google/gemma-4-26b-a4b-qat runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Open run page →

Frequently asked questions

Is google/gemma-4-26b-a4b-qat good for coding?: In our benchmarks, google/gemma-4-26b-a4b-qat scores 65.4/100 for coding. It runs at about 66.9 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 google/gemma-4-26b-a4b-qat good for agentic (tool-using) tasks?: In our benchmarks, google/gemma-4-26b-a4b-qat scores 61.9/100 for agentic workflows. It runs at about 66.9 tok/s, so if you want more speed, swift-iq2_xs is faster (~91.5 tok/s) and still scores well for agentic workflows (91.4/100).
How fast is google/gemma-4-26b-a4b-qat for local inference?: Across 8 community benchmark runs, google/gemma-4-26b-a4b-qat reaches up to 157.0 tok/s and averages 66.9 tok/s, with the fastest results on NVIDIA GeForce RTX 4090.
How much memory does google/gemma-4-26b-a4b-qat need?: The leanest observed configuration used about 18.5 GB of memory (quantizations tested: 4bit, Q4_0).
Which tools have been used to run google/gemma-4-26b-a4b-qat?: Benchmarks were submitted using LM Studio. Results are community-contributed and updated as new runs arrive.