Is google/gemma-4-e2b good for coding?

In our benchmarks, google/gemma-4-e2b scores 46.1/100 for coding. It runs at about 64.2 tok/s, so if you want more speed, ornith-ai/Ornith-1.5-35B-A3B-GGUF is faster (~80.3 tok/s) and still scores well for coding (82.1/100).

Is google/gemma-4-e2b good for agentic (tool-using) tasks?

In our benchmarks, google/gemma-4-e2b scores 69.8/100 for agentic workflows. It runs at about 64.2 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-e2b for local inference?

Across 25 community benchmark runs, google/gemma-4-e2b reaches up to 106.2 tok/s and averages 64.2 tok/s, with the fastest results on NVIDIA GeForce RTX 3060 Laptop GPU.

How much memory does google/gemma-4-e2b need?

The leanest observed configuration used about 4.6 GB of memory (quantizations tested: 4bit, 6bit, 8bit, Q4_K_M).

Which tools have been used to run google/gemma-4-e2b?

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

Model benchmarks

google/gemma-4-e2b local LLM performance

As of November 2026, google/gemma-4-e2b runs at up to 106.2 tok/s for local inference (best of 25 community benchmark runs across 5 GPUs).

BionicLM Studio4bit6bit8bitQ4_K_M

ShareReddit X

Model size

Peak speed

106.2 tok/s

Average speed

64.2 tok/s

Avg PP

1586.2 tok/s

Min memory

4.6 GB

Max context

65,536 tokens

Avg output / run

8,167 tokens

Avg runtime / run

4m 42s

Avg quality

62.9

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 25 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	57.5	62.9	68.2
Agent Workflow	57.3	69.8	85.0
Code Generation	22.3	46.1	59.8
Role Play & Narrative	68.9	77.4	85.0
Research & Analysis	38.2	58.1	68.0

Performance by hardware and tool

Every hardware/tool/quantization combination google/gemma-4-e2b 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 3060 Laptop GPU	Bionic	—	106.2 tok/s	105.4 tok/s	6.3 GB	32,768 tokens	62.4	4
Apple M1 Max	LM Studio	4bit	78.9 tok/s	78.6 tok/s	5.3 GB	65,536 tokens	60.4	8
Apple M1 Max	LM Studio	6bit	76.4 tok/s	76.4 tok/s	6.6 GB	8,192 tokens	68.2	1
Apple M1 Max	LM Studio	8bit	74.4 tok/s	65.6 tok/s	7.2 GB	65,536 tokens	66.3	2
Apple M1 Max	LM Studio	Q4_K_M	71.3 tok/s	71.3 tok/s	4.6 GB	32,768 tokens	61.8	1
NVIDIA RTX PRO 500 Blackwell Generation Laptop GPU	LM Studio	—	57.1 tok/s	46.7 tok/s	n/a	8,192 tokens	63.2	3
AMD Radeon RX 6600	LM Studio	—	41.9 tok/s	41.2 tok/s	n/a	8,192 tokens	65.5	2
AMD Radeon Vega Series / Radeon Vega Mobile Series	Bionic	—	20.2 tok/s	20.2 tok/s	7.5 GB	16,384 tokens	60.8	2
CPU only	LM Studio	Q4_K_M	8.4 tok/s	6.9 tok/s	5.9 GB	8,192 tokens	67.2	2

Benchmark runs

All 25 google/gemma-4-e2b 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-e2b good for coding?: In our benchmarks, google/gemma-4-e2b scores 46.1/100 for coding. It runs at about 64.2 tok/s, so if you want more speed, ornith-ai/Ornith-1.5-35B-A3B-GGUF is faster (~80.3 tok/s) and still scores well for coding (82.1/100).
Is google/gemma-4-e2b good for agentic (tool-using) tasks?: In our benchmarks, google/gemma-4-e2b scores 69.8/100 for agentic workflows. It runs at about 64.2 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-e2b for local inference?: Across 25 community benchmark runs, google/gemma-4-e2b reaches up to 106.2 tok/s and averages 64.2 tok/s, with the fastest results on NVIDIA GeForce RTX 3060 Laptop GPU.
How much memory does google/gemma-4-e2b need?: The leanest observed configuration used about 4.6 GB of memory (quantizations tested: 4bit, 6bit, 8bit, Q4_K_M).
Which tools have been used to run google/gemma-4-e2b?: Benchmarks were submitted using Bionic, LM Studio. Results are community-contributed and updated as new runs arrive.