Best local LLMs for MacBook Pro 14 M6 32GB (ships Nov 6 2026)

Best for Coding · 16k context

llama-server -hf unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q4_K_M -c 16384 -fa on --jinja -ngl 99

OLLAMA_CONTEXT_LENGTH=16384 ollama serve
ollama run hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q4_K_M

Context length is an Ollama server setting: quit the Ollama app or stop the service first, or set it in its environment (launchctl on Mac, systemctl on Linux).

pip install -U mlx-lm
mlx_lm.generate --model mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit --prompt "Hello" --max-tokens 256
mlx_lm.server --model mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit --port 8080

MLX runs on Apple Silicon only. Repo: mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit (converted with mlx-lm, per its model card). These commands use MLX's default context handling; there is no MLX speed on this site yet.

Best for Chat / general · 16k context

llama-server -hf unsloth/gemma-4-12b-it-GGUF:Q4_K_M -c 16384 -fa on --jinja -ngl 99

OLLAMA_CONTEXT_LENGTH=16384 ollama serve
ollama run hf.co/unsloth/gemma-4-12b-it-GGUF:Q4_K_M

Context length is an Ollama server setting: quit the Ollama app or stop the service first, or set it in its environment (launchctl on Mac, systemctl on Linux).

MLX: no MLX conversion yet for this model (none verified in mlx-community), so use llama.cpp or ollama above.

Highest-ranked pick here with an MLX conversion: Qwen3-Coder 30B-A3B (Q4_K_M in GGUF).

pip install -U mlx-lm
mlx_lm.generate --model mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit --prompt "Hello" --max-tokens 256
mlx_lm.server --model mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit --port 8080

MLX runs on Apple Silicon only. Repo: mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit (converted with mlx-lm, per its model card). These commands use MLX's default context handling; there is no MLX speed on this site yet.

Best for Vision · 16k context

llama-server -hf unsloth/gemma-4-12b-it-GGUF:Q4_K_M -c 16384 -fa on --jinja -ngl 99

OLLAMA_CONTEXT_LENGTH=16384 ollama serve
ollama run hf.co/unsloth/gemma-4-12b-it-GGUF:Q4_K_M

Context length is an Ollama server setting: quit the Ollama app or stop the service first, or set it in its environment (launchctl on Mac, systemctl on Linux).

MLX: no MLX conversion yet for this model (none verified in mlx-community), so use llama.cpp or ollama above.

Highest-ranked pick here with an MLX conversion: Gemma 4 26B-A4B (UD-Q5_K_M in GGUF).

pip install -U mlx-vlm
mlx_vlm.generate --model mlx-community/gemma-4-26b-a4b-it-4bit --prompt "Hello" --max-tokens 256
mlx_vlm.server --model mlx-community/gemma-4-26b-a4b-it-4bit

MLX runs on Apple Silicon only. Repo: mlx-community/gemma-4-26b-a4b-it-4bit (converted with mlx-vlm, per its model card). These commands use MLX's default context handling; there is no MLX speed on this site yet.

All 19 models that fit, fastest first · 16k context

Click a column header to sort. 8 of 19 speeds here are measured or scaled from measured; the rest are bandwidth estimates. Filter by use case, speed source, fit, license and more in the interactive advisor →

Long context on this hardware · up to 1M tokens

Largest context whose KV cache still fits next to the weights (same placement as the 16k result, KV types f16 / q8_0 / q4_0). Models are capped at their verified maximum; extended means beyond the native maximum using YaRN, only listed where the model card documents it. Long context needs flash attention (-fa on, already in the commands) and a lot of memory; a quantized KV cache saves memory but can cost accuracy. Speed at depth is an estimate (penalty fitted to measured depth runs up to 128k).

Try it in the advisor: 128k · 256k · 512k · 1M

MLX on this Mac

MLX is Apple's machine-learning framework for Apple Silicon. Where an mlx-community conversion exists, the commands above include an MLX block next to llama.cpp and ollama.

• File size. MLX's default quantization is affine with a group size of 64, and every group stores a scale and a bias (MLX docs). With 16-bit scales and biases that works out to about 4.5 bits per weight at 4-bit and 8.5 at 8-bit. llama.cpp lists Q4_K_M at about 4.89 and Q8_0 at about 8.50 bits per weight (llama.cpp quantize README). So an MLX 4-bit repo is usually a little smaller than Q4_K_M and MLX 8-bit is about the same size as Q8_0. Real repos differ because some layers (embeddings, vision towers, MoE experts in MXFP4 for gpt-oss) are stored differently, so we show each repo's actual size.
• Quality. They are different schemes. MLX 4-bit applies one bit width with per-group scale and bias, while Q4_K_M is a K-quant mix that keeps some tensors at higher precision (llama.cpp k-quants PR). We have not found a rigorous like-for-like quality comparison, so we only pair them by bit class: MLX 4-bit with Q4-class GGUF and MLX 8-bit with Q8_0.
• Speed. MLX speeds are not shown on this site yet. Published comparisons often find MLX faster than llama.cpp on Macs: an M2 Ultra study measured about 230 tok/s for MLX against about 150 for llama.cpp (short context) on Qwen-2.5 models (arXiv 2511.05502), and an M4 Max table shows mlx-lm ahead of llama.cpp on most 4-bit text models but behind on Gemma 3 4B (arXiv 2601.19139). So MLX may well run faster than the llama.cpp numbers shown here, but not for every model, and we do not estimate by how much.
• Which package. Text-only conversions run with mlx-lm. Many recent mlx-community repos (Gemma 4, Qwen3.5/3.6/3.8, Qwen3-VL) were converted with mlx-vlm and their model cards say to run them with it, so the command we show follows each repo's card.

Repos are checked against the Hugging Face API (how).

Related

Models that fit fully, fastest first: LFM3 1.7B · GLM-4.7-Flash · Laguna XS 2.1 · Qwen3-Coder 30B-A3B · LFM2.5 24B-A2B.

Other hardware with a similar memory size: 32 GB tier · NVIDIA RTX Spark N1X 128GB · all hardware · leaderboard

Measured speed pages: gpt-oss 20B · Qwen3.5 9B (all)

FAQ