Rankings

Explore benchmarks

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

Clear filters

Top Engines

Top Engines, Desktop, Speed score.
#EngineSpeed score out of 100
198 /100 Score
289 /100 Score
386 /100 Score
Not ranked in this view
Not rankedMagnitude Magnitude Unsupported
Not rankeduzu Mirai Unsupported
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, model-dtype. Its weights are 4-bit in another scheme: mlx-affine-4bit-g64, where most use Q4_K_M. 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: f16, model-dtype.

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

Magnitude: Unsupported in 2 comparisons. Magnitude 0.2.17 reports this GGUF as Unsupported (magnitude catalog inspect)

uzu: Unsupported in 2 comparisons. uzu 0.6.17 renders its own Qwen3.8 chat template: 67 input tokens for the 27-token canonical check prompt, so it can't run the frozen prompts (measured in qualification).

Coverage

Speed score per device, out of 100: each device ranked on its own.
Desktop
M5 Max 64 GiB macOS 27 Apple M5 Max
llama.cpp86 out of 100, in 2 of 2 comparisons
MLX LM98 out of 100, in 2 of 2 comparisons
Ollama89 out of 100, in 2 of 2 comparisons
MagnitudeUnsupported Unsupported in 2 comparisons. Magnitude 0.2.17 reports this GGUF as Unsupported (magnitude catalog inspect)
uzuUnsupported Unsupported in 2 comparisons. uzu 0.6.17 renders its own Qwen3.8 chat template: 67 input tokens for the 27-token canonical check prompt, so it can't run the frozen prompts (measured in qualification).

Speed

TTFT ms

Generation tok/s

Speed, Qwen3.8 27B on M5 Max · 64 GiB, 4-bit · Archive 2048 input / 128 output · macOS 27
RankEngineDecode and TTFT
1MLX LM, approximate comparison, see details Apple (ml-explore)33.1 tok/s 2,145 ms TTFT
2Ollama, approximate comparison, see details Ollama25.9 tok/s 2,823 ms TTFT
3llama.cpp, approximate comparison, see details ggml-org25.1 tok/s 3,092 ms TTFT
Not rankedMagnitude MagnitudeUnsupported
Not rankeduzu MiraiUnsupported
details

Qwen3.8 27B on M5 Max · 64 GiB · 4-bit · 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, model-dtype. Its weights are 4-bit in another scheme: mlx-affine-4bit-g64, where most use Q4_K_M.

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

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

Magnitude: Magnitude 0.2.17 reports this GGUF as Unsupported (magnitude catalog inspect)

uzu: uzu 0.6.17 renders its own Qwen3.8 chat template: 67 input tokens for the 27-token canonical check prompt, so it can't run the frozen prompts (measured in qualification).

Memory usage by engine

Memory usage by engine, Qwen3.8 27B on M5 Max · 64 GiB, 4-bit · Archive 2048 input / 128 output · macOS 27
RankEngineMemory
1MLX LM, approximate comparison, see details Apple (ml-explore)17.87 GiB
2Ollama, approximate comparison, see details Ollama19.07 GiB
3llama.cpp, approximate comparison, see details ggml-org19.08 GiB
Not rankedMagnitude MagnitudeUnsupported
Not rankeduzu MiraiUnsupported
details

Qwen3.8 27B on M5 Max · 64 GiB · 4-bit · Archive 2048 input / 128 output · macOS 27

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

MLX LM: † Approximate comparison. The experiments use different effective KV cache precision: f16, model-dtype. Its weights are 4-bit in another scheme: mlx-affine-4bit-g64, where most use Q4_K_M.

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

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

Magnitude: Magnitude 0.2.17 reports this GGUF as Unsupported (magnitude catalog inspect)

uzu: uzu 0.6.17 renders its own Qwen3.8 chat template: 67 input tokens for the 27-token canonical check prompt, so it can't run the frozen prompts (measured in qualification).

Every comparison

Comparisons in this view: each engine's score in each comparison, out of 100, or why it has none.
ComparisonMLX LMOllamallama.cppMagnitudeuzu
Qwen3.8 27B on M5 Max · 64 GiB 4-bit · Archive 2048 input / 128 output · macOS 27 100, approximate78, approximate76, approximateUnsupportedUnsupported
Qwen3.8 27B on M5 Max · 64 GiB 8-bit · Archive 2048 input / 128 output · macOS 27 97, approximate100, approximate98, approximateUnsupportedUnsupported

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