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 | uzu†, an approximate comparison, see details | 100 /100 Score |
| 2 | Magnitude†, an approximate comparison, see details | 96 /100 Score |
| 3 | Rapid-MLX†, an approximate comparison, see details | 92 /100 Score |
| 4 | MLX LM†, an approximate comparison, see details | 90 /100 Score |
| 5 | Ollama†, an approximate comparison, see details | 77 /100 Score |
| 6 | llama.cpp†, an approximate comparison, see details | 76 /100 Score |
| 7 | LiteRT-LM†, an approximate comparison, see details | 26 /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
† 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
| 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 | 761/1 out of 100, in 1 of 1 comparisons | 1001/1 out of 100, in 1 of 1 comparisons | 811/1 out of 100, in 1 of 1 comparisons | 691/1 out of 100, in 1 of 1 comparisons |
| LiteRT-LM | 261/1 out of 100, in 1 of 1 comparisons | 761/1 out of 100, in 1 of 1 comparisons | 1001/1 out of 100, in 1 of 1 comparisons | Not measured on this device. |
| uzu | 1001/1 out of 100, in 1 of 1 comparisons | Not measured on this device. | Not measured on this device. | 1001/1 out of 100, in 1 of 1 comparisons |
| Magnitude | 961/1 out of 100, in 1 of 1 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
| MLX LM | 901/1 out of 100, in 1 of 1 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
| Ollama | 771/1 out of 100, in 1 of 1 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
| Rapid-MLX | 921/1 out of 100, in 1 of 1 comparisons | Not measured on this device. | Not measured on this device. | Not measured on this device. |
Speed
| Rank | Engine | Decode and TTFT |
|---|---|---|
| 1 | uzu†, approximate comparison, see details | 776.1 tok/s 88.8 ms TTFT |
| 2 | Magnitude†, approximate comparison, see details | 747.9 tok/s 44.4 ms TTFT |
| 3 | Rapid-MLX†, approximate comparison, see details | 712.2 tok/s 61.9 ms TTFT |
| 4 | MLX LM†, approximate comparison, see details | 702.2 tok/s 57.7 ms TTFT |
| 5 | Ollama†, approximate comparison, see details | 599.8 tok/s 61.5 ms TTFT |
| 6 | llama.cpp†, approximate comparison, see details | 592.1 tok/s 63.4 ms TTFT |
| 7 | LiteRT-LM†, approximate comparison, see details | 198.9 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
| Rank | Engine | Memory |
|---|---|---|
| 1 | Ollama†, approximate comparison, see details | 0.61 GiB |
| 2 | llama.cpp†, approximate comparison, see details | 0.61 GiB |
| 3 | uzu†, approximate comparison, see details | 0.70 GiB |
| 4 | Magnitude†, approximate comparison, see details | 1.04 GiB |
| 5 | LiteRT-LM†, approximate comparison, see details | 1.25 GiB |
| 6 | Rapid-MLX†, approximate comparison, see details | 2.05 GiB |
| 7 | MLX LM†, approximate comparison, see details | 2.09 GiB |
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
| Comparison | uzu | Magnitude | Rapid-MLX | MLX LM | Ollama | llama.cpp | LiteRT-LM |
|---|---|---|---|---|---|---|---|
| LFM2.5 350M on M5 Max · 64 GiB 8-bit · Archive 2048 input / 128 output · macOS 27 | 100†, approximate | 96†, approximate | 92†, approximate | 90†, approximate | 77†, approximate | 76†, approximate | 26†, approximate |
Each engine’s Speed score in each matched comparison, out of 100, or why it has none. How rankings work