Run LFM2.5-2.6B on M5 Pro

LFM2.5-2.6B runs at 156 tok/s (156.16 as recorded) on an M5 Pro, a published figure reported by huggingface.co (MLX, 4-bit, decode at 4K context depth). It needs about 2 GB of unified memory. It is Liquid AI's 2.6B dense model, released under the LFM Open licence.

Speed
sourced156
tok/s
First Token
—
seconds
RAM Needed
2
GB minimum
Engine
MLX
of this figure
Want to run LFM2.5-2.6B faster — or step up to bigger models? Find your Mac →

The figure

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.

MetricValue
Tokens per secondsourced156 tok/s
Provenancepublished figure · huggingface.co
Quantization4-bit
EngineMLX
Prompt processing (prefill)sourced4102 tok/s at 4K context
Contextdecode at 4K context depth
Figure dated2026-10-19
Minimum RAM2 GB
Parameters2.6B
Chip bandwidth307 GB/s

Every figure for LFM2.5-2.6B on the M5 Pro

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 (2 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.

SpeedQuantEngineMac RAMContextSourceNote
sourced156.16 tok/s4-bitMLX—decode at 4K context depthhuggingface.coCompetitor baseline on SparkLLM's Spark-X2.6-1.7B card chart 'Mac M5 Pro · 16-core' (the 16-core-GPU M5 Pro has the same 307 GB/s): stock MLX-LM 0.32.2 on an INT4 mlx-community checkpoint. 120.74 tok/s at 32K.
Provenance key
sourced
Published figure — vendor or press review
estimated
community
Linked individual run — blog, forum or submission

Speed by context depth: LFM2.5-2.6B on the M5 Pro

Measured figures only, each from its source’s own session; the index does not estimate long-context or prefill speed, so a missing cell is —. Each @ column is plain decode with that many tokens already in the prompt, from the same session as the row’s short-context figure: decode slows as the KV cache it reads grows. Prefill is prompt processing, the tokens per second a runtime reads a prompt at; with long agent contexts it sets how long you wait before the first token.

RunDecode (row)@4K@32K@128KPrefill
MLX · 4-bit · huggingface.cosourced156.16 tok/s
decode at 4K context depth
sourced156.16 tok/ssourced120.74 tok/ssourced68.04 tok/ssourced4102 tok/s
at 4K

Setup: run LFM2.5-2.6B on a Mac

Ollama — the index has not verified an Ollama tag for LFM2.5-2.6B, 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:

pip install -U mlx-lm
mlx_lm.generate --model <repo-id> --prompt "Hello"

Downloads: Search the Ollama library → · Hugging Face model card →

The 156 tok/s figure above came from MLX at 4-bit; 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.

LFM2.5-2.6B on other Apple Silicon chips

ChipSpeedQuantEngineBandwidth
M5 Maxsourced220 tok/sQ4_K_MMLX614 GB/s
M4 Maxestimated203 tok/sQ4_K_M—546 GB/s
M6estimated79 tok/sQ4_K_MMLX170 GB/s

For the estimated figures here, the bandwidth formula’s typical error for dense models is ±9%, against 75 measured runs of 20 models (how it is computed).

About LFM2.5-2.6B

Liquid AI's 2.6B model, ~220 tok/s on M5 Max. LFM Open license, LLMCheck capability 14/50.

LFM Open licence · 2.6B · 131K context · LLMCheck capability 14/50 · see its leaderboard rank →

Memory by context length

Estimated, not measured: the index’s minimum for LFM2.5-2.6B (2 GB — the weights at the quant that fits, plus ~2 GB) plus the KV cache for that many tokens, computed from its config.json. hybrid short-conv: attention on 8 of 30 layers; conv layers keep a constant state. A 16-bit KV cache is the default in llama.cpp, Ollama, LM Studio and mlx-lm; a q8 cache halves the KV part. Coding agents hold 64K tokens or more. Runtimes also add a buffer that grows with context, so leave a few GB spare. macOS lets the GPU use only part of a Mac’s memory by default (figures for an M5 Pro Mac, which runs macOS 26 or later): “fits” means within the GPU memory a Mac of that tier can use with the limit raised, a tier tagged raise GPU limit fits only after sudo sysctl iogpu.wired_limit_mb raises it (the tooltip has the exact command), and the last column is the smallest Mac memory size whose default limit holds it (the GPU memory limit).

ContextKV cache (16-bit)RAM neededRAM, q8 KVSmallest tier that fitsWithin the default GPU limit
8K~0.1 GB~2.1 GB~2.0 GB8 GB Macsame
32K~0.5 GB~2.5 GB~2.2 GB8 GB Macsame
64K (coding-agent context)~1.1 GB~3.0 GB~2.5 GB8 GB Macsame
128K~2.1 GB~4.1 GB~3.0 GB8 GB Macsame

System requirements

Macs with the M5 Pro

🛒 Get a Mac that runs LFM2.5-2.6B

LFM2.5-2.6B needs about 2 GB of unified memory. These Apple Silicon Macs have enough to run it.

As an Amazon Associate, LLMCheck earns from qualifying purchases. These affiliate links cost you nothing extra and help keep the LLMCheck index free.

Compare More Models

See how LFM2.5-2.6B stacks up against other models on your specific Mac hardware.

Open Compare Tool Full Leaderboard