Rankings

Explore benchmarks

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

Top Engines

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

† uzu: 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: mirai-l, where most use Q8_0.

† Magnitude: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its phases were timed by the harness clock around the engine API, so they include transport that engine counters leave out.

† Rapid-MLX: 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.

† 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.

† Ollama: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default.

† llama.cpp: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default.

† 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.

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.cpp76 out of 100, in 1 of 1 comparisons100 out of 100, in 1 of 1 comparisons81 out of 100, in 1 of 1 comparisons69 out of 100, in 1 of 1 comparisons
LiteRT-LM26 out of 100, in 1 of 1 comparisons76 out of 100, in 1 of 1 comparisons100 out of 100, in 1 of 1 comparisonsNot measured on this device.
uzu100 out of 100, in 1 of 1 comparisonsNot measured on this device.Not measured on this device.100 out of 100, in 1 of 1 comparisons
Magnitude96 out of 100, in 1 of 1 comparisonsNot measured on this device.Not measured on this device.Not measured on this device.
MLX LM90 out of 100, in 1 of 1 comparisonsNot measured on this device.Not measured on this device.Not measured on this device.
Ollama77 out of 100, in 1 of 1 comparisonsNot measured on this device.Not measured on this device.Not measured on this device.
Rapid-MLX92 out of 100, in 1 of 1 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 M5 Max · 64 GiB, 8-bit · Archive 2048 input / 128 output · macOS 27
RankEngineDecode and TTFT
1uzu, approximate comparison, see details Mirai776.1 tok/s 88.8 ms TTFT
2Magnitude, approximate comparison, see details Magnitude747.9 tok/s 44.4 ms TTFT
3Rapid-MLX, approximate comparison, see details Rapid-MLX contributors712.2 tok/s 61.9 ms TTFT
4MLX LM, approximate comparison, see details Apple (ml-explore)702.2 tok/s 57.7 ms TTFT
5Ollama, approximate comparison, see details Ollama599.8 tok/s 61.5 ms TTFT
6llama.cpp, approximate comparison, see details ggml-org592.1 tok/s 63.4 ms TTFT
7LiteRT-LM, approximate comparison, see details Google AI Edge198.9 tok/s 635 ms TTFT
details

LFM2.5 350M on M5 Max · 64 GiB · 8-bit · Archive 2048 input / 128 output · macOS 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: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 8-bit in another scheme: mirai-l, where most use Q8_0.

Magnitude: † Approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its phases were timed by the harness clock around the engine API, so they include transport that engine counters leave out.

Rapid-MLX: † 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.

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.

Ollama: † Approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default.

llama.cpp: † Approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default.

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.

Memory usage by engine

details

LFM2.5 350M on M5 Max · 64 GiB · 8-bit · Archive 2048 input / 128 output · macOS 27

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

Ollama: † Approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default.

llama.cpp: † Approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default.

uzu: † 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: mirai-l, where most use Q8_0.

Magnitude: † Approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its phases were timed by the harness clock around the engine API, so they include transport that engine counters leave out.

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.

Rapid-MLX: † 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.

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.

Every comparison

Comparisons in this view: each engine's score in each comparison, out of 100, or why it has none.
ComparisonuzuMagnitudeRapid-MLXMLX LMOllamallama.cppLiteRT-LM
LFM2.5 350M on M5 Max · 64 GiB 8-bit · Archive 2048 input / 128 output · macOS 27 100, approximate96, approximate92, approximate90, approximate77, approximate76, approximate26, approximate

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