Skip to content
Local AI Wizard
Your machine
Theme

Mac mini M2 · 24 GB unified

Apple · Mac mini · 2023

125 text models run well · 100 GB/s bandwidth

Models on this hardware

Start here

Text

Within 24 GB: 125 run well · 2 tight · 23 reviewed · 17 more with 32 GB

Text · LMArena

rating · higher is better

  • Color = publisher

Source: LMArena 2026-10-22 · collected 2026-10-25 · dataset · CC BY 4.0 · Models without a rating are left out, not scored as zero.

Rows run well first, then tight fits; within each, reviewed models, then Hub downloads over 30 days, not quality. Sort a score column to rank by that source alone. No model runs only partly on this Mac: it either fits in memory or it does not.
Text models that run well within 24 GB
gpt-oss-20bOpenAI · 21B · LMArena 1,317Q4 · published params · source1,317 ± 6~22 GBTight fit: Needs ~22 GB: runs well, about 2 GB left beyond the system reserveQ4 · published paramssource
Granite 4.2 3BIBM · 3B · LMArena 1,290Q4 · published params · source1,290 ± 12~9 GBQ4 · published paramssource
Granite 4.2 8BIBM · 8B · LMArena 1,290Q4 · published params · source1,290 ± 11~12 GBQ4 · published paramssource
Phi-4Microsoft · 14B · LMArena 1,256Q4 · published params · source1,256 ± 5~17 GBQ4 · published paramssource
Llama 3.1 8B InstructMeta · 8B · LMArena 1,211Q4 · published params · source1,211 ± 4~12 GBQ4 · published paramssource
Llama 3.2 3B InstructMeta · 3.21B · LMArena 1,167Q4 · published params · source1,167 ± 8~9 GBQ4 · published paramssource
Llama 3.2 1B InstructMeta · 1.23B · LMArena 1,111Q4 · published params · source1,111 ± 8~7 GBQ4 · published paramssource
Qwen3 14BAlibaba · 14.8BQ4 · published params · sourceNot on the LMArena text leaderboard of 2026-10-22~18 GBQ4 · published paramssource
8 of 127

No model within 24 GB has a LiveBench score, so the table has no LiveBench column. LMArena ratings in this table: LMArena 2026-10-22 · collected 2026-10-25 · dataset · CC BY 4.0

Tight fit: it runs, but leaves less than 2.4 GB free beyond the 6 GB the system keeps (the memory figure already includes that reserve).

Every reviewed text model here 23

Every one that runs here, well or slowly, scored or not; 7 of 23 have a LiveBench or LMArena result. Run well first, then tight fits, then slow; within each, by Hub downloads over 30 days, read 2026-10-25, not by quality.

Reviewed text models that run on Mac mini M2 · 24 GB unified
Qwen3 14BAlibaba · 14.8BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~18 GB9.0Apache 2.010.4M
Qwen3 8BAlibaba · 8.2BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~13 GB5.0Apache 2.08.1M
Qwen3.5 9BAlibaba · 9BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~13 GB5.7Apache 2.06.1M
Llama 3.1 8B InstructMeta · 8BNot in LiveBench 2026-06-251,211~12 GB4.9Llama 3.14.8M
Qwen3 1.7BAlibaba · 1.7BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~8 GB1.1Apache 2.04.1M
Gemma 4 E4BGoogle · 8BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~12 GB5.0Apache 2.04M
Llama 3.2 1B InstructMeta · 1.2BNot in LiveBench 2026-06-251,111~7 GB0.8Llama 3.23.5M
Gemma 4 E2BGoogle · 5.1BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~10 GB3.1Apache 2.03M
8 of 23

With 32 GB you also run… 17

The next memory size on this ladder runs these well. Figures are the requirement on this hardware type.

6 of 17

Browse Text models

Adjust the estimate: quantization, context, reserveloads the live catalog

Vision

Within 24 GB: 125 run well · 4 reviewed · 13 more with 32 GB

Vision · LMArena

rating · higher is better

  • Color = publisher

Source: LMArena 2026-10-19 · collected 2026-10-25 · dataset · CC BY 4.0 · Models without a rating are left out, not scored as zero.

Rows run well first, then tight fits; within each, reviewed models, then Hub downloads over 30 days, not quality. Sort a score column to rank by that source alone. No model runs only partly on this Mac: it either fits in memory or it does not.
Vision models that run well within 24 GB
Molmo2-8Ballenai · 8.7B · Image to text · LMArena 1,112Q4 · Hugging Face count · source1,112 ± 20~13 GBQ4 · Hugging Face countsource
Qwen2-VL-7B-InstructQwen · 8.29B · Image to text · LMArena 1,031Q4 · Hugging Face count · source1,031 ± 11~13 GBQ4 · Hugging Face countsource
Molmo-7B-D-0924allenai · 8B · Image to text · LMArena 994Q4 · Hugging Face count · source994 ± 14~12 GBQ4 · Hugging Face countsource
Llama-3.2-11B-Vision-Instructmeta-llama · 10.7B · Image to text · LMArena 993Q4 · Hugging Face count · source993 ± 11~15 GBQ4 · Hugging Face countsource
MiniCPM-V-2_6openbmb · 8.1B · Image to text · LMArena 962Q4 · Hugging Face count · source962 ± 16~13 GBQ4 · Hugging Face countsource
Phi-3.5-vision-instructmicrosoft · 4.1B · Image to text · LMArena 919Q4 · Hugging Face count · source919 ± 16~10 GBQ4 · Hugging Face countsource
Gemma 4 12BGoogle · 11.95B · Image to textQ4 · published params · sourceNot on the LMArena vision leaderboard of 2026-10-19~15 GBQ4 · published paramssource
Qwen3.5 9BAlibaba · 9B · Image to textQ4 · published params · sourceNot on the LMArena vision leaderboard of 2026-10-19~13 GBQ4 · published paramssource
8 of 125

LMArena ratings in this table: LMArena 2026-10-19 · collected 2026-10-25 · dataset · CC BY 4.0

With 32 GB you also run… 13

The next memory size on this ladder runs these well. Figures are the requirement on this hardware type.

6 of 13

12 more vision models in the catalog have no usable memory figure or parameter count for this profile. Browse Vision models

Image

50 within 24 GB · 10 more with 32 GB

Image · Prompt alignment · GenEval

0–1 · higher is better

Source: GenEval · 2023 · reviewed 2026-10-05 · Published quality in the original benchmark preset; not measured on this hardware.

Image models within 24 GB
Stable Diffusion XL 1.0Stability AI · 2.6B · Text to imagePlatform path documented · Apple Core ML · source0.55Not on the LMArena text-to-image leaderboard of 2026-10-07Documented path without a memory figurePlatform path documentedApple Core ML · source
Stable Diffusion 1.5Stability AI · 0.98B · Text to imagePlatform path documented · Apple Core ML · source0.43Not on the LMArena text-to-image leaderboard of 2026-10-07Documented path without a memory figurePlatform path documentedApple Core ML · source
FLUX.2 [klein] 9BBlack Forest Labs · 9B · Text to image · LMArena 1,07016-bit est. · params · sourceNo comparable published result1,070 ± 3~24 GB16-bit est. · paramssource
FLUX.2 [klein] Base 9BBlack Forest Labs · 9B · Text to image16-bit est. · params · sourceNo comparable published resultArena rated flux-2-klein-9b; this entry is the undistilled Base checkpoint~24 GB16-bit est. · paramssource
FLUX.2-klein-9b-kvblack-forest-labs · 9.1B · Image to image16-bit est. · params · sourceNo comparable published resultNot reviewed for LMArena~24 GB16-bit est. · paramssource
HiDream-O1-Image-Dev-2604HiDream-ai · 8.8B · Text + image to image16-bit est. · params · sourceNo comparable published resultNot reviewed for LMArena~24 GB16-bit est. · paramssource
ERNIE ImageBaidu · 8B · Text to image16-bit est. · params · sourceNo comparable published resultNot on the LMArena text-to-image leaderboard of 2026-10-07~22 GB16-bit est. · paramssource
ERNIE Image TurboBaidu · 8.03B · Text to image16-bit est. · params · sourceNo comparable published resultNot on the LMArena text-to-image leaderboard of 2026-10-07~22 GB16-bit est. · paramssource
8 of 50

LMArena ratings in this table: LMArena 2026-10-07 · collected 2026-10-25 · dataset · CC BY 4.0

With 32 GB you also run… 10

The next memory size on this ladder. Figures are the requirement on this hardware type.

6 of 10

17 more image models in the catalog have no usable memory figure or parameter count for this profile. Browse Image models

Video

29 within 24 GB · 2 more with 32 GB
Video models within 24 GB
HunyuanVideo-1.5tencent · 8.3B · Text to video16-bit est. · params · source~22 GB16-bit est. · paramssource
CogVideoX-5b-I2Vzai-org · 5.6B · Image to video16-bit est. · params · source~16 GB16-bit est. · paramssource
CogVideoX1.5-5Bzai-org · 5.6B · Text to video16-bit est. · params · source~16 GB16-bit est. · paramssource
CogVideoX1.5-5B-I2Vzai-org · 5.6B · Image to video16-bit est. · params · source~16 GB16-bit est. · paramssource
CogVideoX-5BZ.ai / Tsinghua KEG · 5B · Text to video16-bit est. · params · source~14 GB16-bit est. · paramssource
Wan2.2-TI2V-5BWan-AI · 5B · Text to video16-bit est. · params · source~14 GB16-bit est. · paramssource
Wan2.2-TI2V-5B-DiffusersWan-AI · 5B · Text to video16-bit est. · params · source~14 GB16-bit est. · paramssource
SANA-Video_2.0_5B_720pEfficient-Large-Model · 4.5B · Text to video16-bit est. · params · source~13 GB16-bit est. · paramssource
8 of 29

How this list works Documented execution paths on this hardware first, then publisher minimums (CUDA or Apple MPS).

With 32 GB you also run… 2

The next memory size on this ladder. Figures are the requirement on this hardware type.

21 more video models in the catalog have no usable memory figure or parameter count for this profile. Browse Video models

Audio

128 within 24 GB · 2 more with 32 GB

Audio · Word error rate · Open ASR

% · lower is better

Source: Word error rate · Open ASR · reviewed 2026-10-20 · Published quality in the original benchmark preset; not measured on this hardware.

Audio models within 24 GB
Whisper large-v3-turboOpenAI · 0.81B · Speech to textPlatform path documented · whisper.cpp / Metal · source7.83Documented path without a memory figurePlatform path documentedwhisper.cpp / Metal · source
Whisper large-v3OpenAI · 1.55B · Speech to textPlatform path documented · whisper.cpp / Metal · source7.44Documented path without a memory figurePlatform path documentedwhisper.cpp / Metal · source
Llasa-8BHKUSTAudio · 8.6B · Text to speech16-bit est. · params · sourceNo comparable published result~23 GB16-bit est. · paramssource
MOSS-TTS-v1.5OpenMOSS-Team · 8.5B · Text to speech16-bit est. · params · sourceNo comparable published result~23 GB16-bit est. · paramssource
personaplex-7b-v1nvidia · 8.4B · Speech to speech / audio to audio16-bit est. · params · sourceNo comparable published result~23 GB16-bit est. · paramssource
Qwen2-Audio-7B-InstructQwen · 8.4B · Speech to text16-bit est. · params · sourceNo comparable published result~23 GB16-bit est. · paramssource
VibeVoice-ASRmicrosoft · 8.7B · Speech to text16-bit est. · params · sourceNo comparable published result~23 GB16-bit est. · paramssource
VibeVoice-ASR-Streaming-7Bmicrosoft · 8.7B · Speech to text16-bit est. · params · sourceNo comparable published result~23 GB16-bit est. · paramssource
8 of 128

With 32 GB you also run… 2

The next memory size on this ladder. Figures are the requirement on this hardware type.

35 more audio models in the catalog have no usable memory figure or parameter count for this profile. Browse Audio models

Embeddings

90 run well within 24 GB

Embeddings · MTEB (Multilingual, v2) · mean(task)

0–100 · higher is better

  • Color = publisher

Source: MTEB results (curated rows, sourced per model)

Embeddings models that run well within 24 GB
Qwen3-Embedding 8BAlibaba (Qwen)8-bit · published params · source70.6~12 GB8-bit · published paramssource
Qwen3-Embedding 4BAlibaba (Qwen)8-bit · published params · source69.5~7 GB8-bit · published paramssource
Harrier OSS v1 (0.6B)Microsoft8-bit · published params · source69.0~3 GB8-bit · published paramssource
Qwen3-Embedding 0.6BAlibaba (Qwen)8-bit · published params · source64.3~3 GB8-bit · published paramssource
Multilingual E5 Large Instructintfloat8-bit · published params · source63.2~3 GB8-bit · published paramssource
EmbeddingGemma 2Google8-bit · published params · source61.4~3 GB8-bit · published paramssource
EmbeddingGemma (300M)Google8-bit · published params · source61.1~3 GB8-bit · published paramssource
BGE-M3BAAI8-bit · published params · source59.6~3 GB8-bit · published paramssource
8 of 90

Browse Embeddings models

Runs well within 24 GB of unified memory · planning estimates, not measured speed bundled catalog copy · 1,777 repositories · checked

Measurements

Published tokens per second, exactly as measured; never estimated.

Published measurements on this hardware
Llama 7B Q4_0 (llama-bench reference model)Not named by the summary table; llama-bench…Apple M2 · 10-core GPU · 100 GB/s (memory size not stated by the source)21.91Q4_0
0 / 128 · Single llama-bench process;…
Metal
llama.cpp commit 8e672ef
llama.cpp Apple Silicon summary · community result
2026-10-25

Specifications

21 sourced fields; the source is named once per group, and a field from another source links to it. Official specifications

Unified memory24 GBShared by CPU and GPU · source
Memory bandwidth100 GB/sManufacturer spec · not model speed · source

Marker: the lightest text model that does not run well here but runs well with 32 GB. No text model that runs well here has a LiveBench score. Q4 planning estimate, not a measured fit.

Selected configuration

Latest check · Manufacturer source

Memory capacity
24 GB
Memory type
24 GB unified
Chip / processor
Apple M2 · 8-core CPU / 10-core GPU
Memory bandwidth
100 GB/s
CPU
Apple M2
CPU cores
8
GPU
Apple M2 GPU
GPU cores
10
Memory type
LPDDR5

Family specification

Latest check · Manufacturer source

Introduced
2023
Storage options (family)
256 GB · 512 GB · 1 TB · 2 TB SSD
Ports
2 × Thunderbolt 4 · 2 × USB-A · HDMI · Ethernet · 3.5 mm
Maximum continuous power (specification)
150 W
Wi-Fi
Wi-Fi 6E (802.11ax)
Bluetooth
Bluetooth 5.3
Dimensions (cm)
19.70 × 19.70 × 3.58 (W × D × H)
Neural Engine cores
16
Operating system
macOS
Maximum Thunderbolt rate
40 Gb/s
Weight
1.18 kg
Release date
2023-01-24

Apple lists 24 GB as the maximum configurable unified memory for the base M2 mini; its chip has an 8-core CPU, 10-core GPU and 100 GB/s bandwidth. The OS and apps share the memory, and Apple publishes no local AI speed result for this profile.

Photo: Official manufacturer image. · Manufacturer source

Memory bandwidth is not measured model speed; verify runtime and checkpoint compatibility. Hardware specification reviewed: 2026-10-13.

How estimates work
Candidates
Reviewed pages use published parameter counts; automatic text candidates use official repository tensor counts (packed quantized weights and unsupported architectures excluded).
Planning
Q4 weights + 20% allowance + reserve: 25% of unified or system memory (at least 2 GB, at most 8 GB; 6 GB here) or 2 GB per dedicated GPU. Adjustable in “Adjust the estimate”. No KV cache is calculated from context.
Fit tiers
Runs well: the estimate fits the fast memory (VRAM, or unified memory within the macOS GPU share). Runs slowly: within the machine's memory but outside the fast part; a graphics card offloads layers to PC RAM (the listed RAM, or 32 GB assumed). Doesn't run: above both. Tiers are planning estimates, never a speed.
Vision
Vision-language models use the Q4 planning estimate from the parameter count (published by the model page, or read from Hugging Face). Classic vision models (classify, detect, segment) use a 16-bit estimate. No vision benchmark is tracked, so rows carry no score.
Media and embeddings
Media rows use documented execution paths (exact hardware or platform) and publisher minimums; a published requirement above this memory moves a model to the next tier. Without either, a 16-bit estimate from the published parameter count (× 1.2 + 2 GB), labelled as such. Embeddings use 8-bit weights + 20% + 2 GB.

Local AI basics: parameters, Q4 and memory · Full model catalog · Data and methodology

Alternatives

Other machines on the same memory ladder (unified memory); each page lists the models that fit it.