Explore benchmarks
Filter by model, hardware, workload and quantization. Every score links to the comparisons and experiments behind it.
Top Engines
| # | Engine | Speed score out of 100 |
|---|---|---|
| 1 | MLX LM†, an approximate comparison, see details | 99 /100 Score |
| ||
| 2 | Rapid-MLX†, an approximate comparison, see details | 99 /100 Score |
| ||
| 3 | Magnitude†, an approximate comparison, see details | 98 /100 Score |
| ||
| 4 | Ollama†, an approximate comparison, see details | 96 /100 Score |
| ||
| 5 | llama.cpp†, an approximate comparison, see details | 91 /100 Score |
| ||
| 6 | ZML†, an approximate comparison, see details | 61 /100 Score |
| ||
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: 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.
† 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.
† 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.
† 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
| Desktop | Mobile | |||
|---|---|---|---|---|
| M5 Max 64 GiB macOS 27 Apple M5 Max | OnePlus 13 16 GB Android 16 Qualcomm Snapdragon 8 Elite (SM8750) | Pixel 8 8 GB Android 16 Google Tensor G3 | iPhone 12 4 GB iOS 27 Apple A14 Bionic | |
| llama.cpp | 913/3 out of 100, in 3 of 3 comparisons | No rival yet No other engine has a result in the same comparison yet. | Failed Failed in 2 comparisons. No trial was valid. | No rival yet No other engine has a result in the same comparison yet. |
| Magnitude | 982/3 out of 100, in 2 of 3 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
| MLX LM | 992/3 out of 100, in 2 of 3 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
| Ollama | 963/3 out of 100, in 3 of 3 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
| Rapid-MLX | 992/3 out of 100, in 2 of 3 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
| ZML | 612/3 out of 100, in 2 of 3 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
Speed
| Rank | Engine | Prefill |
|---|---|---|
| 1 | MLX LM†, approximate comparison, see details | 21,872 tok/s |
| 2 | Rapid-MLX†, approximate comparison, see details | 20,564 tok/s |
| 3 | Magnitude†, approximate comparison, see details | 18,058 tok/s |
| 4 | ZML†, approximate comparison, see details | 15,773 tok/s |
| 5 | Ollama†, approximate comparison, see details | 15,385 tok/s |
| 6 | llama.cpp†, approximate comparison, see details | 14,523 tok/s |
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.
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.
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.
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.
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.
Memory usage by engine
| Rank | Engine | Memory |
|---|---|---|
| 1 | Ollama†, approximate comparison, see details | 2.45 GiB |
| 2 | llama.cpp†, approximate comparison, see details | 2.45 GiB |
| 3 | Magnitude†, approximate comparison, see details | 2.96 GiB |
| 4 | MLX LM†, approximate comparison, see details | 3.26 GiB |
| 5 | Rapid-MLX†, approximate comparison, see details | 3.45 GiB |
| 6 | ZML†, approximate comparison, see details | 4.75 GiB |
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
| Comparison | MLX LM | Rapid-MLX | Magnitude | Ollama | llama.cpp | ZML |
|---|---|---|---|---|---|---|
| LFM2.5 1.2B Instruct on M5 Max · 64 GiB bf16 · Archive 2048 input / 128 output · macOS 27 | 99†, approximate | 98†, approximate | 100†, approximate | 99†, approximate | 92†, approximate | 76†, approximate |
| LFM2.5 2.6B on M5 Max · 64 GiB bf16 · Archive 2048 input / 128 output · macOS 27 | Not run | Not run | Unsupported | 100 | 97 | Unsupported |
| LFM2.5 350M on M5 Max · 64 GiB bf16 · Archive 2048 input / 128 output · macOS 27 | 99†, approximate | 100†, approximate | 96†, approximate | 88†, approximate | 83†, approximate | 49†, approximate |
Each engine’s Speed score in each matched comparison, out of 100, or why it has none. How rankings work