Benchmark result
gemma4:e2b on NVIDIA RTX 3500 Ada Generation Laptop GPU — 149.6 tok/s
Measured with Ollama 0.40.15 on November 12, 2026.
Support llm-bench.io
llm-bench.io is free. If it helped you, I'd appreciate your support to cover the infrastructure costs.
What the model built
Open full screen →The coding scenario asks for a playable game in a single HTML file. This is exactly what the model returned, unedited; the frame only adds a line that tells this page how tall it is.
How this run compares
35th fastest of 36 runs of this model on NVIDIA RTX 3500 Ada Generation Laptop GPU · median 153.3 tok/s.
sort
- 1158.5tok/s73.4Ollama · Q4_K_M
- 2157.9tok/s71.6Ollama · Q4_K_M
- 3157.8tok/s69.2Ollama · Q4_K_M
- 4157.6tok/s68.2Ollama · Q4_K_M
- 5156.6tok/s64.4Ollama · Q4_K_M
- 6156.3tok/s64.5Ollama · Q4_K_M
- 7156.1tok/s68.7Ollama · Q4_K_M
- 8155.4tok/s68.2Ollama · Q4_K_M
- 9155.4tok/s61.3Ollama · Q4_K_M
- 10155.1tok/s69.0Ollama · Q4_K_M
- 11154.9tok/s61.4Ollama · Q4_K_M
- 12154.8tok/s63.1Ollama · Q4_K_M
- 13154.8tok/s65.7Ollama · Q4_K_M
- 14154.4tok/s67.7Ollama · Q4_K_M
- 15154.2tok/s71.0Ollama · Q4_K_M
- 16154.2tok/s74.0Ollama · Q4_K_M
- 17154.2tok/s60.9Ollama · Q4_K_M
- 18153.4tok/s69.5Ollama · Q4_K_M
- 19153.3tok/s74.7Ollama · Q4_K_M
- 20152.7tok/s64.3Ollama · Q4_K_M
- 21152.7tok/s69.0Ollama · Q4_K_M
- 22152.6tok/s70.3Ollama · Q4_K_M
- 23152.2tok/s71.8Ollama · Q4_K_M
- 24151.7tok/s73.2Ollama · Q4_K_M
- 25151.5tok/s67.0Ollama · Q4_K_M
- 26151.5tok/s73.4Ollama · Q4_K_M
- 27151.4tok/s69.3Ollama · Q4_K_M
- 28151.1tok/s64.9Ollama · Q4_K_M
- 29151.0tok/s73.6Ollama · Q4_K_M
- 30150.9tok/s66.6Ollama · Q4_K_M
- 31150.8tok/s58.7Ollama · Q4_K_M
- 32150.8tok/s73.7Ollama · Q4_K_M
- 33150.4tok/s65.6Ollama · Q4_K_M
- 34150.4tok/s69.8Ollama · Q4_K_M
- 35149.6tok/s69.4Ollama · Q4_K_Mthis run
- 36143.2tok/s68.2Ollama · Q4_K_M
gemma4:e2b on other hardware
Reproduce this run
toolOllama 0.40.15
modelgemma4:e2b (Q4_K_M)
context16,384
thinkingon
samplingtemperature 1 · top_p 0.95 · top_k 64
clientv0.4.99+102
Set those in Ollama, then:
llm-benchmark benchmark --model "gemma4:e2b" --tool "Ollama"The client prompts for context length and thinking mode, and for the KV cache dtype on Unsloth Studio. Sampling is left at the model default — the values above are what the tool reported using, not overrides the client sent.