Model benchmarks
gemma-4-E4B-it-M-TS-Q4_K_M local LLM performance
As of November 2026, gemma-4-E4B-it-M-TS-Q4_K_M runs at up to 83.8 tok/s for local inference (best of 5 community benchmark runs across 1 GPU).
Model size
4B
Peak speed
83.8 tok/s
Average speed
81.1 tok/s
Avg PP
1536.5 tok/s
Min memory
5.2 GB
Max context
16,384 tokens
Avg output / run
8,682 tokens
Avg runtime / run
1m 51s
Avg quality
70.9
Benchmark runs
5
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 5 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 | 66.8 | 70.9 | 76.1 |
| Agent Workflow | 70.0 | 80.4 | 90.1 |
| Code Generation | 55.9 | 61.7 | 68.0 |
| Role Play & Narrative | 71.7 | 78.4 | 86.0 |
| Research & Analysis | 58.0 | 63.3 | 68.1 |
Performance by hardware and tool
Every hardware/tool/quantization combination gemma-4-E4B-it-M-TS-Q4_K_M 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 | llama.cpp | Q4_K_M | 83.8 tok/s | 81.1 tok/s | 5.2 GB | 16,384 tokens | 70.9 | 5 |
Benchmark runs
All 5 gemma-4-E4B-it-M-TS-Q4_K_M runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is gemma-4-E4B-it-M-TS-Q4_K_M good for coding?
- In our benchmarks, gemma-4-E4B-it-M-TS-Q4_K_M scores 61.7/100 for coding. It runs at about 81.1 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 gemma-4-E4B-it-M-TS-Q4_K_M good for agentic (tool-using) tasks?
- In our benchmarks, gemma-4-E4B-it-M-TS-Q4_K_M scores 80.4/100 for agentic workflows. It runs at about 81.1 tok/s, so if you want more speed, qwen3.8-flash-next-iq3_s is faster (~98.1 tok/s) and still scores well for agentic workflows (90.6/100).
- How fast is gemma-4-E4B-it-M-TS-Q4_K_M for local inference?
- Across 5 community benchmark runs, gemma-4-E4B-it-M-TS-Q4_K_M reaches up to 83.8 tok/s and averages 81.1 tok/s, with the fastest results on NVIDIA GeForce RTX 3060.
- How much memory does gemma-4-E4B-it-M-TS-Q4_K_M need?
- The leanest observed configuration used about 5.2 GB of memory (quantizations tested: Q4_K_M).
- Which tools have been used to run gemma-4-E4B-it-M-TS-Q4_K_M?
- Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.