Rankings

Explore benchmarks

Filter by model, hardware, workload and quantization. Every score links to the comparisons and experiments behind it.

Clear filters

Top Engines

Top Engines, Overall, Speed score.
#EngineSpeed score out of 100
195 /100 Score
290 /100 Score
387 /100 Score
456 /100 Score
Not ranked in this view
Not rankedRapid-MLX Rapid-MLX contributors Too few results
Not rankeduzu Mirai Too few results
Not rankedMagnitude Magnitude Too few results
details

Speed Decode speed: tokens generated per second.

Score: 100 is the best result in every comparison an engine entered; 80 is 80% of the best, on average. Only complete experiments with a passed output check count. How rankings work

The score combines counted comparisons. Select one comparison to see its measured values.

† MLX LM: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 8-bit in another scheme: mlx-affine-8bit-g64, where most use Q8_0. The experiments use different effective KV cache precision: litert-lm default, f16, model-dtype. The experiments here use different 8-bit schemes: litert-int8, Q8_0, mlx-affine-8bit-g64. The experiments use different effective KV cache precision: f16, model-dtype.

† Ollama: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. The experiments use different effective KV cache precision: litert-lm default, f16, model-dtype. The experiments here use different 8-bit schemes: litert-int8, Q8_0, mlx-affine-8bit-g64. The experiments use different effective KV cache precision: f16, model-dtype.

† llama.cpp: approximate comparison. The experiments use different effective KV cache precision: f16, uzu default. The experiments here use different 8-bit schemes: Q8_0, mirai-l. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. The experiments use different effective KV cache precision: litert-lm default, f16, model-dtype. The experiments here use different 8-bit schemes: litert-int8, Q8_0, mlx-affine-8bit-g64. The experiments use different effective KV cache precision: f16, model-dtype. The experiments use different effective KV cache precision: litert-lm default, f16. The experiments here use different 8-bit schemes: litert-int8, Q8_0.

† LiteRT-LM: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 8-bit in another scheme: litert-int8, where most use Q8_0. The experiments use different effective KV cache precision: litert-lm default, f16. The experiments here use different 8-bit schemes: litert-int8, Q8_0.

Rapid-MLX: Results in 3 of 8 comparisons; a rank needs 4.

uzu: Results in 3 of 8 comparisons; a rank needs 4.

Magnitude: Results in 2 of 8 comparisons; a rank needs 4.

Coverage

Speed score per device, out of 100: each device ranked on its own.
DesktopMobile
M5 Max 64 GiB macOS 27 Apple M5 MaxOnePlus 13 16 GB Android 16 Qualcomm Snapdragon 8 Elite (SM8750)Pixel 8 8 GB Android 16 Google Tensor G3iPhone 12 4 GB iOS 27 Apple A14 Bionic
llama.cpp87 out of 100, in 4 of 4 comparisons100 out of 100, in 2 of 2 comparisons81 out of 100, in 1 of 1 comparisons69 out of 100, in 1 of 1 comparisons
LiteRT-LM29 out of 100, in 2 of 4 comparisons81 out of 100, in 2 of 2 comparisons100 out of 100, in 1 of 1 comparisonsNot measured on this device.
uzu100 out of 100, in 2 of 4 comparisonsNot measured on this device.Not measured on this device.100 out of 100, in 1 of 1 comparisons
Magnitude93 out of 100, in 2 of 4 comparisonsNot measured on this device.Not measured on this device.Not measured on this device.
MLX LM95 out of 100, in 4 of 4 comparisonsNot measured on this device.Not measured on this device.Not measured on this device.
Ollama90 out of 100, in 4 of 4 comparisonsNot measured on this device.Not measured on this device.Not measured on this device.
Rapid-MLX95 out of 100, in 3 of 4 comparisonsNot measured on this device.Not measured on this device.Not measured on this device.

Speed

TTFT ms

Generation tok/s

Speed, LFM2.5 350M on iPhone 12 · 4 GB, 8-bit · Archive 2048 input / 128 output · iOS 27
RankEngineDecode and TTFT
1uzu, approximate comparison, see details Mirai70.2 tok/s 2,356 ms TTFT
2llama.cpp, approximate comparison, see details ggml-org48.3 tok/s 2,210 ms TTFT
details

LFM2.5 350M on iPhone 12 · 4 GB · 8-bit · Archive 2048 input / 128 output · iOS 27

Medians with the model warmed; model loading is excluded. Each table uses one matched comparison.

uzu: † Approximate comparison. The experiments use different effective KV cache precision: f16, uzu default. The experiments here use different 8-bit schemes: Q8_0, mirai-l.

llama.cpp: † Approximate comparison. The experiments use different effective KV cache precision: f16, uzu default. The experiments here use different 8-bit schemes: Q8_0, mirai-l.

Memory usage by engine

Memory usage by engine, LFM2.5 350M on iPhone 12 · 4 GB, 8-bit · Archive 2048 input / 128 output · iOS 27
RankEngineMemory
Not rankedllama.cpp ggml-orgNot measured
Not rankeduzu MiraiNot measured
details

LFM2.5 350M on iPhone 12 · 4 GB · 8-bit · Archive 2048 input / 128 output · iOS 27

Memory ranks require comparable, calibrated measurements. Unavailable measurements remain unranked.

llama.cpp: No peak memory measured. † Approximate comparison. The experiments use different effective KV cache precision: f16, uzu default. The experiments here use different 8-bit schemes: Q8_0, mirai-l.

uzu: No peak memory measured. † Approximate comparison. The experiments use different effective KV cache precision: f16, uzu default. The experiments here use different 8-bit schemes: Q8_0, mirai-l.

Every comparison

Comparisons in this view: each engine's score in each comparison, out of 100, or why it has none.
ComparisonMLX LMOllamallama.cppLiteRT-LMRapid-MLXuzuMagnitude
LFM2.5 350M on iPhone 12 · 4 GB 8-bit · Archive 2048 input / 128 output · iOS 27 Not runNot run69, approximateNot runNot run100, approximateNot run
LFM2.5 1.2B Instruct on M5 Max · 64 GiB 8-bit · Archive 2048 input / 128 output · macOS 27 92, approximate87, approximate87, approximate32, approximate92, approximate100, approximate89, approximate
LFM2.5 2.6B on M5 Max · 64 GiB 8-bit · Archive 2048 input / 128 output · macOS 27 100, approximate96, approximate89, approximateNot counted100, approximateUnsupportedUnsupported
LFM2.5 350M on M5 Max · 64 GiB 8-bit · Archive 2048 input / 128 output · macOS 27 90, approximate77, approximate76, approximate26, approximate92, approximate100, approximate96, approximate
Qwen3.8 27B on M5 Max · 64 GiB 8-bit · Archive 2048 input / 128 output · macOS 27 97, approximate100, approximate98, approximateNot runNot runUnsupportedUnsupported
LFM2.5 1.2B Instruct on OnePlus 13 · 16 GB 8-bit · Archive 2048 input / 128 output · Android 16 Not runNot run100, approximate87, approximateNot runNot runNot run
LFM2.5 350M on OnePlus 13 · 16 GB 8-bit · Archive 2048 input / 128 output · Android 16 Not runNot run100, approximate76, approximateNot runNot runNot run
LFM2.5 350M on Pixel 8 · 8 GB 8-bit · Archive 2048 input / 128 output · Android 16 Not runNot run81, approximate100, approximateNot runNot runNot run
LFM2.5 1.2B Instruct on iPhone 12 · 4 GB 8-bit · Archive 2048 input / 128 output · iOS 27 Not counted: fewer than two scored enginesNot runNot runMeasured, approximateNot runNot runNot countedNot run
LFM2.5 2.6B on iPhone 12 · 4 GB 8-bit · Archive 2048 input / 128 output · iOS 27 Not counted: fewer than two scored enginesNot runNot runUnsupportedNot runNot runUnsupportedNot run
LFM2.5 2.6B on OnePlus 13 · 16 GB 8-bit · Archive 2048 input / 128 output · Android 16 Not counted: fewer than two scored enginesNot runNot runMeasured, approximateNot countedNot runNot runNot run
LFM2.5 1.2B Instruct on Pixel 8 · 8 GB 8-bit · Archive 2048 input / 128 output · Android 16 Not counted: fewer than two scored enginesNot runNot runMeasured, approximateNot countedNot runNot runNot run
LFM2.5 2.6B on Pixel 8 · 8 GB 8-bit · Archive 2048 input / 128 output · Android 16 Not counted: fewer than two scored enginesNot runNot runFailedFailedNot runNot runNot run

Each engine’s Speed score in each matched comparison, out of 100, or why it has none. How rankings work