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

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

† MLX LM: approximate comparison. The experiments use different effective KV cache precision: f16, magnitude default, model-dtype, zml default.

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

† Rapid-MLX: approximate comparison. The experiments use different effective KV cache precision: f16, magnitude default, model-dtype, zml default.

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

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

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.cpp92 out of 100, in 1 of 1 comparisonsFailed Failed in 1 comparison. No trial was valid.Failed Failed in 1 comparison. No trial was valid.Unsupported Unsupported in 1 comparison. llama.cpp errored on the opposite check in qualification: The app wrote no result for run (exit 1); iOS ends an app that exceeds its memory limit or leaves the foreground.
Magnitude100 out of 100, in 1 of 1 comparisonsNot measured on this device.Not measured on this device.Not measured on this device.
MLX LM99 out of 100, in 1 of 1 comparisonsNot measured on this device.Not measured on this device.Not measured on this device.
Ollama99 out of 100, in 1 of 1 comparisonsNot measured on this device.Not measured on this device.Not measured on this device.
Rapid-MLX98 out of 100, in 1 of 1 comparisonsNot measured on this device.Not measured on this device.Not measured on this device.
ZML76 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 1.2B Instruct on M5 Max · 64 GiB, bf16 · Archive 2048 input / 128 output · macOS 27
RankEngineDecode and TTFT
1Magnitude, approximate comparison, see details Magnitude199.6 tok/s 113 ms TTFT
2MLX LM, approximate comparison, see details Apple (ml-explore)198.3 tok/s 94.2 ms TTFT
3Ollama, approximate comparison, see details Ollama197.6 tok/s 136 ms TTFT
4Rapid-MLX, approximate comparison, see details Rapid-MLX contributors195.1 tok/s 100 ms TTFT
5llama.cpp, approximate comparison, see details ggml-org184.2 tok/s 142 ms TTFT
6ZML, approximate comparison, see details ZML150.9 tok/s 130 ms TTFT
details

LFM2.5 1.2B Instruct on M5 Max · 64 GiB · bf16 · Archive 2048 input / 128 output · macOS 27

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

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

MLX LM: † Approximate comparison. The experiments use different effective KV cache precision: f16, magnitude default, model-dtype, zml default.

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

Rapid-MLX: † Approximate comparison. The experiments use different effective KV cache precision: f16, magnitude default, model-dtype, zml default.

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

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

Memory usage by engine

details

LFM2.5 1.2B Instruct on M5 Max · 64 GiB · bf16 · 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: f16, magnitude default, model-dtype, zml default.

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

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

MLX LM: † Approximate comparison. The experiments use different effective KV cache precision: f16, magnitude default, model-dtype, zml default.

Rapid-MLX: † Approximate comparison. The experiments use different effective KV cache precision: f16, magnitude default, model-dtype, zml default.

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

Every comparison

Comparisons in this view: each engine's score in each comparison, out of 100, or why it has none.
ComparisonMagnitudeMLX LMOllamaRapid-MLXllama.cppZML
LFM2.5 1.2B Instruct on M5 Max · 64 GiB bf16 · Archive 2048 input / 128 output · macOS 27 100, approximate99, approximate99, approximate98, approximate92, approximate76, approximate

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