Explore benchmarks
Filter by model, hardware, workload and quantization. Every score links to the comparisons and experiments behind it.
Top Engines
Overall needs mobile results; each platform is ranked on its own until then.
Desktop
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.
† 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. The experiments use different effective KV cache precision: f16, model-dtype.
† 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. The experiments use different effective KV cache precision: f16, model-dtype.
† 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 | 914/4 out of 100, in 4 of 4 comparisons | No rival yet No other engine has a result in the same comparison yet. | Failed Failed in 2 comparisons. No trial was valid. | Unsupported Unsupported in 2 comparisons. 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. |
| Magnitude | 1002/4 out of 100, in 2 of 4 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
| MLX LM | 1004/4 out of 100, in 4 of 4 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
| Ollama | 992/4 out of 100, in 2 of 4 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
| Rapid-MLX | 982/4 out of 100, in 2 of 4 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
| ZML | 852/4 out of 100, in 2 of 4 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
Speed
| Rank | Engine | Decode and TTFT |
|---|---|---|
| 1 | Ollama†, approximate comparison, see details | 199.6 tok/s 33.8 ms TTFT |
| 2 | Magnitude†, approximate comparison, see details | 199.6 tok/s 28.7 ms TTFT |
| 3 | MLX LM†, approximate comparison, see details | 199.0 tok/s 30.7 ms TTFT |
| 4 | Rapid-MLX†, approximate comparison, see details | 198.0 tok/s 32.7 ms TTFT |
| 5 | ZML†, approximate comparison, see details | 192.6 tok/s 24.9 ms TTFT |
| 6 | llama.cpp†, approximate comparison, see details | 190.5 tok/s 36.7 ms TTFT |
details
LFM2.5 1.2B Instruct on M5 Max · 64 GiB · bf16 · Archive 256 input / 128 output · macOS 27
Medians with the model warmed; model loading is excluded. Each table uses one matched comparison.
Ollama: † 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.
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 | 2.43 GiB |
| 2 | llama.cpp†, approximate comparison, see details | 2.43 GiB |
| 3 | MLX LM†, approximate comparison, see details | 2.44 GiB |
| 4 | Rapid-MLX†, approximate comparison, see details | 2.70 GiB |
| 5 | Magnitude†, approximate comparison, see details | 2.95 GiB |
| 6 | ZML†, approximate comparison, see details | 4.93 GiB |
details
LFM2.5 1.2B Instruct on M5 Max · 64 GiB · bf16 · Archive 256 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.
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.
Every comparison
| Comparison | llama.cpp | Magnitude | MLX LM | Ollama | Rapid-MLX | ZML |
|---|---|---|---|---|---|---|
| LFM2.5 1.2B Instruct on M5 Max · 64 GiB bf16 · Archive 2048 input / 128 output · macOS 27 | 92†, approximate | 100†, approximate | 99†, approximate | 99†, approximate | 98†, approximate | 76†, approximate |
| LFM2.5 1.2B Instruct on M5 Max · 64 GiB bf16 · Archive 256 input / 128 output · macOS 27 | 95†, approximate | 100†, approximate | 100†, approximate | 100†, approximate | 99†, approximate | 96†, approximate |
| LFM2.5 1.2B Instruct on M5 Max · 64 GiB bf16 · 2048 input / 128 output · macOS 27 | 92†, approximate | Not run | 100†, approximate | Not run | Not run | Not run |
| LFM2.5 1.2B Instruct on M5 Max · 64 GiB bf16 · 256 input / 128 output · macOS 27 | 85†, approximate | Not run | 100†, approximate | Not run | Not run | Not run |
| LFM2.5 1.2B Instruct on iPhone 12 · 4 GB bf16 · Archive 2048 input / 128 output · iOS 27 Not counted: fewer than two scored engines | Unsupported | Not run | Not run | Not run | Not run | Not run |
| LFM2.5 1.2B Instruct on iPhone 12 · 4 GB bf16 · Archive 256 input / 128 output · iOS 27 Not counted: fewer than two scored engines | Unsupported | Not run | Not run | Not run | Not run | Not run |
| LFM2.5 1.2B Instruct on OnePlus 13 · 16 GB bf16 · Archive 2048 input / 128 output · Android 16 Not counted: fewer than two scored engines | Failed | Not run | Not run | Not run | Not run | Not run |
| LFM2.5 1.2B Instruct on OnePlus 13 · 16 GB bf16 · Archive 256 input / 128 output · Android 16 Not counted: fewer than two scored engines | Measured | Not run | Not run | Not run | Not run | Not run |
| LFM2.5 1.2B Instruct on Pixel 8 · 8 GB bf16 · Archive 2048 input / 128 output · Android 16 Not counted: fewer than two scored engines | Failed | Not run | Not run | Not run | Not run | Not run |
| LFM2.5 1.2B Instruct on Pixel 8 · 8 GB bf16 · Archive 256 input / 128 output · Android 16 Not counted: fewer than two scored engines | Failed | Not run | Not run | Not run | Not run | Not run |
Each engine’s Speed score in each matched comparison, out of 100, or why it has none. How rankings work