5 builds of Gemma 4 12B measured on real machines — different packagers, precisions and engines, each a different file with its own numbers. The best each engine proved is first; every build has its own page with the machines and signatures behind it. 12B parameters, as the vendor states.
trained 256K · proved 256K context proved by unsloth Q8_0
29.1 tok/s peak, by unsloth Q4_K_M
proved by at least one build: Tool calling · Tool loop finished · Faithful to tool results · Chained tool calls · Structured output · Sustained output · Extended thinking · Vision · Counts in images · Multiple images at once · Spatial reasoning · Writes code · Reasons about code · Synthesises sources · Follows instructions
| build | publisher | precision | engine | whose work | context proved | best tok/s | machines |
|---|---|---|---|---|---|---|---|
| hf.co/unsloth/gemma-4-12b-it-GGUF:Q8_0 | unsloth | Q8_0 | ollama | quantizer | 2,62,144 | 18.6 | 1 |
| hf.co/unsloth/gemma-4-12b-it-GGUF:Q4_K_M | unsloth | Q4_K_M | ollama | quantizer | 1,31,072 | 29.1 | 1 |
| hf.co/google/gemma-4-12B-it-qat-q4_0-gguf:Q4_0 | Q4_0 | ollama | vendor | 1,31,072 | 27.9 | 1 | |
| hf.co/unsloth/gemma-4-12b-it-GGUF:IQ4_XS | unsloth | IQ4_XS | ollama | quantizer | 1,31,072 | 23.2 | 1 |
| hf.co/unsloth/gemma-4-12b-it-GGUF:BF16 | unsloth | BF16 | ollama | quantizer | 1,31,072 | 10.7 | 1 |