10 builds of Phi 4 MINI-REASONING 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.
proved 4K context proved by unsloth IQ3_XXS
189.4 tok/s peak, by unsloth IQ3_XXS
proved by at least one build: Structured output · Sustained output · Extended thinking · Reasons about code · Synthesises sources
| build | publisher | precision | engine | whose work | context proved | best tok/s | machines |
|---|---|---|---|---|---|---|---|
| hf.co/unsloth/Phi-4-mini-reasoning-GGUF:IQ3_XXS | unsloth | IQ3_XXS | ollama | quantizer | 4,096 | 189.4 | 6 |
| hf.co/lmstudio-community/Phi-4-mini-reasoning-GGUF:Q8_0 | lmstudio-community | Q8_0 | ollama | — | 4,096 | 22.2 | 1 |
| hf.co/unsloth/Phi-4-mini-reasoning-GGUF:IQ1_S | unsloth | IQ1_S | ollama | quantizer | — | 188.9 | 9 |
| hf.co/lmstudio-community/Phi-4-mini-reasoning-GGUF:Q3_K_L | lmstudio-community | Q3_K_L | ollama | — | — | 89.9 | 4 |
| hf.co/unsloth/Phi-4-mini-reasoning-GGUF:Q2_K | unsloth | Q2_K | ollama | quantizer | — | 37.7 | 3 |
| hf.co/unsloth/Phi-4-mini-reasoning-GGUF:Q3_K_M | unsloth | Q3_K_M | ollama | quantizer | — | 31.9 | 4 |
| hf.co/lmstudio-community/Phi-4-mini-reasoning-GGUF:Q4_K_M | lmstudio-community | Q4_K_M | ollama | — | — | 7.3 | 2 |
| hf.co/unsloth/Phi-4-mini-reasoning-GGUF:Q2_K_L | unsloth | Q2_K_L | ollama | quantizer | — | 6.8 | 3 |
| hf.co/unsloth/Phi-4-mini-reasoning-GGUF:Q4_K_M | unsloth | Q4_K_M | ollama | quantizer | — | 6.5 | 2 |
| hf.co/lmstudio-community/Phi-4-mini-reasoning-GGUF:Q6_K | lmstudio-community | Q6_K | ollama | — | — | — | 1 |