Spark-X2.6-4B runs at 76 tok/s (76.14 as recorded) on an M5 Pro, a published figure reported by huggingface.co (llama.cpp, Q4_K_M, decode at 4K context depth). It needs about 4 GB of unified memory. It is SparkLLM (XHToken)'s 4.11B dense model, released under the Apache 2.0 licence.
Want to run Spark-X2.6-4B faster — or step up to bigger models? Find your Mac →This is a published figure: a real run, reported by its source and linked below. The index records it as given and never adjusts it.
| Metric | Value |
|---|---|
| Tokens per second | sourced76 tok/s |
| Provenance | published figure · huggingface.co |
| Quantization | Q4_K_M |
| Engine | llama.cpp |
| Context | decode at 4K context depth |
| Figure dated | 2026-10-21 |
| Minimum RAM | 4 GB |
| Parameters | 4.11B |
| Chip bandwidth | 307 GB/s |
Each row as the index holds it, with its source. “Mac RAM” is the memory of the Mac a run used, not the model’s requirement (4 GB). The headline figure follows the reference rule: the lower-median measured plain-decode run at 5 bits or fewer (a run whose quant was not stated counts only when no stated-quant run qualifies), else the estimate; a run marked “not used to anchor estimates” is listed here but never becomes the headline.
| Speed | Quant | Engine | Mac RAM | Context | Source | Note |
|---|---|---|---|---|---|---|
| sourced86.58 tok/s | INT4 | MLX | — | decode at 4K context depth | huggingface.co | SparkLLM's card chart 'Mac M5 Pro · 16-core': MLX INT4 through SparkLLM's own Spark-MLX-LLM loader, so it does not anchor. 72.05 tok/s at 32K.; not used to anchor estimates |
| sourced76.14 tok/s | Q4_K_M | llama.cpp | — | decode at 4K context depth | huggingface.co | SparkLLM's card chart 'Mac M5 Pro · 16-core' (the 16-core-GPU M5 Pro has the same 307 GB/s); llama.cpp build not stated. 62.04 tok/s at 32K. |
Ollama — the index has not verified an Ollama tag for Spark-X2.6-4B, so it doesn’t print one (a guessed tag fails with “pull model manifest: file does not exist”). Search the Ollama library →
MLX — pick a build from Hugging Face’s MLX listings, then run it with its repo id:
Downloads: Hugging Face repo → · model card →
The 76 tok/s figure above came from llama.cpp at Q4_K_M; other runtimes and quants run at different speeds.
Prefer a GUI? LM Studio downloads GGUF and MLX builds in one click. Step by step: install Ollama on a Mac.
| Chip | Speed | Quant | Engine | Bandwidth |
|---|---|---|---|---|
| M5 Max | estimated158 tok/s | Q4_K_M | llama.cpp | 614 GB/s |
SparkLLM's (XHToken) Apache 2.0 4.1B dense on-device agent model (21 Oct 2026): 1M context, thinking levels none/low/high. Its card's vendor-run table claims SWE-bench Verified 62.8 and GPQA Diamond 77.1, figures not independently reproduced, with no scaffold stated. Stock llama.cpp, mlx-lm 0.32+ and LM Studio run it; a 4-bit build is ~2.3–2.6 GB, but Ollama's only tag is the 8.2 GB full-precision file. SparkLLM's card lists 76 tok/s decode on an M5 Pro (llama.cpp Q4_K_M, 4K context).
Apache 2.0 licence · 4.11B · 1000K context · LLMCheck capability 20/50 · see its leaderboard rank →
Spark-X2.6-4B needs about 4 GB of unified memory. These current Apple Silicon Macs have the headroom to run it comfortably:
As an Amazon Associate, LLMCheck earns from qualifying purchases. These affiliate links cost you nothing extra and help keep the LLMCheck index free.
See how Spark-X2.6-4B stacks up against other models on your specific Mac hardware.
Open Compare Tool Full Leaderboard