Is google/gemma-4-e4b good for coding?

In our benchmarks, google/gemma-4-e4b scores 46.6/100 for coding. It runs at about 42.1 tok/s, so if you want more speed, nvidia/Qwen3.8-27B-NVFP4 is faster (~63.3 tok/s) and still scores well for coding (82.7/100).

Is google/gemma-4-e4b good for agentic (tool-using) tasks?

In our benchmarks, google/gemma-4-e4b scores 72.9/100 for agentic workflows. It runs at about 42.1 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-e4b for local inference?

Across 10 community benchmark runs, google/gemma-4-e4b reaches up to 78.6 tok/s and averages 42.1 tok/s, with the fastest results on NVIDIA GeForce RTX 3070.

How much memory does google/gemma-4-e4b need?

The leanest observed configuration used about 4.0 GB of memory (quantizations tested: 4bit, Q4_K_M, Q8_0).

Which tools have been used to run google/gemma-4-e4b?

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

Model benchmarks

google/gemma-4-e4b local LLM performance

As of November 2026, google/gemma-4-e4b runs at up to 78.6 tok/s for local inference (best of 10 community benchmark runs across 6 GPUs).

LM Studiollama.cpp4bitQ4_K_MQ8_0

ShareReddit X

Model size

Peak speed

78.6 tok/s

Average speed

42.1 tok/s

Avg PP

718.5 tok/s

Min memory

4.0 GB

Max context

128,000 tokens

Avg output / run

9,694 tokens

Avg runtime / run

4m 47s

Avg quality

63.4

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 10 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	51.4	63.4	71.0
Agent Workflow	63.6	72.9	80.8
Code Generation	30.7	46.6	58.4
Role Play & Narrative	62.7	75.1	81.1
Research & Analysis	23.4	59.1	71.7

Performance by hardware and tool

Every hardware/tool/quantization combination google/gemma-4-e4b has been benchmarked on, ranked by peak token generation speed. Last updated November 2026.

Hardware	Tool	Quant	Best tok/s	Avg tok/s	Memory	Context	Quality	Runs
NVIDIA GeForce RTX 3070	llama.cpp	—	78.6 tok/s	78.6 tok/s	4.0 GB	8,192 tokens	69.6	1
AMD Radeon RX 6800/6800 XT / 6900 XT	LM Studio	Q4_K_M	65.5 tok/s	65.5 tok/s	8.5 GB	32,768 tokens	61.0	1
Apple M1 Max	LM Studio	4bit	48.0 tok/s	46.5 tok/s	7.9 GB	65,536 tokens	63.4	4
NVIDIA GeForce RTX 5070	LM Studio	Q8_0	42.3 tok/s	42.3 tok/s	n/a	32,768 tokens	65.1	1
Advanced Micro Devices, Inc. [AMD/ATI] Phoenix1	LM Studio	—	19.9 tok/s	19.9 tok/s	n/a	32,768 tokens	72.1	1
AMD Radeon 780M Graphics	LM Studio	Q4_K_M	19.7 tok/s	19.7 tok/s	7.0 GB	65,536 tokens	69.5	1
CPU only	LM Studio	—	8.7 tok/s	8.7 tok/s	n/a	128,000 tokens	43.8	1

Benchmark runs

All 10 google/gemma-4-e4b 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-e4b good for coding?: In our benchmarks, google/gemma-4-e4b scores 46.6/100 for coding. It runs at about 42.1 tok/s, so if you want more speed, nvidia/Qwen3.8-27B-NVFP4 is faster (~63.3 tok/s) and still scores well for coding (82.7/100).
Is google/gemma-4-e4b good for agentic (tool-using) tasks?: In our benchmarks, google/gemma-4-e4b scores 72.9/100 for agentic workflows. It runs at about 42.1 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-e4b for local inference?: Across 10 community benchmark runs, google/gemma-4-e4b reaches up to 78.6 tok/s and averages 42.1 tok/s, with the fastest results on NVIDIA GeForce RTX 3070.
How much memory does google/gemma-4-e4b need?: The leanest observed configuration used about 4.0 GB of memory (quantizations tested: 4bit, Q4_K_M, Q8_0).
Which tools have been used to run google/gemma-4-e4b?: Benchmarks were submitted using LM Studio, llama.cpp. Results are community-contributed and updated as new runs arrive.