Apple MacBook Pro 16-inch (2023, M2 Max 38-core GPU): which AI models can it run?

Estimated from the specifications below. Choose the memory size you have or plan to buy, then read the table for that size.

Specifications

Brand
Apple
Year
2023
Chip
Apple M2 Max
Processor
12-core CPU
Graphics
38-core GPU
Memory type
unified memory
Memory options
32 GB, 64 GB, 96 GB
Memory bandwidth
400 GB/s

Sources: https://support.apple.com/en-us/111838

Apple lists 400GB/s and 32GB base memory, configurable to 64GB or 96GB. A 30-core GPU option also exists and is not listed separately.

With 32 GB of memory

About 22.4 GB of it can hold a model.

ModelFitNeeds (GB)Writing speed (tokens per second)SpeedQualityBasis
Llama 3.2 3B
llama3.2:3b
Fits 5 70 to 130 Fast not measured yet Estimated
Qwen3 4B
qwen3:4b · thinks first
Fits 6 56 to 104 Fast not measured yet Estimated
Gemma 3 4B
gemma3:4b
Fits 6.7 42 to 79 Fast not measured yet Estimated
Mistral 7B v0.3
mistral:7b
Fits 7.8 32 to 59 Fast not measured yet Estimated
Llama 3.1 8B
llama3.1:8b
Fits 8.3 29 to 53 Fast not measured yet Estimated
Qwen3 8B
qwen3:8b · thinks first
Fits 8.9 27 to 50 Fast not measured yet Estimated
Gemma 3 12B
gemma3:12b
Fits 15.9 17 to 32 Fast not measured yet Estimated
DeepSeek-R1 Distill Qwen 14B
deepseek-r1:14b · thinks first
Fits 13.7 16 to 29 Fast not measured yet Estimated
Phi-4 14B
phi4:14b
Fits 13.9 15 to 29 Fast not measured yet Estimated
Qwen3 14B
qwen3:14b · thinks first
Fits 13.4 15 to 28 Fast not measured yet Estimated
gpt-oss 20B
gpt-oss:20b · thinks first
Fits 16.5 35 to 65 Fast not measured yet Estimated
Gemma 3 27B
gemma3:27b
Won't fit 27.2 not known n/a not measured yet Estimated
Qwen3 30B (A3B)
qwen3:30b
Won't fit 22.6 not known n/a 9.3/10 Estimated
Qwen3 32B
qwen3:32b
Won't fit 26.3 not known n/a not measured yet Estimated
Llama 3.1 70B
llama3.1:70b
Won't fit 51.5 not known n/a not measured yet Estimated
gpt-oss 120B
gpt-oss:120b
Won't fit 70.5 not known n/a not measured yet Estimated

Models marked "thinks first" can write out their reasoning before they answer, so a job takes longer than the speed alone suggests.

With 64 GB of memory

About 44.8 GB of it can hold a model.

ModelFitNeeds (GB)Writing speed (tokens per second)SpeedQualityBasis
Llama 3.2 3B
llama3.2:3b
Fits 5 70 to 130 Fast not measured yet Estimated
Qwen3 4B
qwen3:4b · thinks first
Fits 6 56 to 104 Fast not measured yet Estimated
Gemma 3 4B
gemma3:4b
Fits 6.7 42 to 79 Fast not measured yet Estimated
Mistral 7B v0.3
mistral:7b
Fits 7.8 32 to 59 Fast not measured yet Estimated
Llama 3.1 8B
llama3.1:8b
Fits 8.3 29 to 53 Fast not measured yet Estimated
Qwen3 8B
qwen3:8b · thinks first
Fits 8.9 27 to 50 Fast not measured yet Estimated
Gemma 3 12B
gemma3:12b
Fits 15.9 17 to 32 Fast not measured yet Estimated
DeepSeek-R1 Distill Qwen 14B
deepseek-r1:14b · thinks first
Fits 13.7 16 to 29 Fast not measured yet Estimated
Phi-4 14B
phi4:14b
Fits 13.9 15 to 29 Fast not measured yet Estimated
Qwen3 14B
qwen3:14b · thinks first
Fits 13.4 15 to 28 Fast not measured yet Estimated
gpt-oss 20B
gpt-oss:20b · thinks first
Fits 16.5 35 to 65 Fast not measured yet Estimated
Gemma 3 27B
gemma3:27b
Fits 27.2 8.2 to 15 Unclear could be Moderate or Fast not measured yet Estimated
Qwen3 30B (A3B)
qwen3:30b · thinks first
Fits 22.6 41 to 76 Fast 9.3/10 Estimated
Qwen3 32B
qwen3:32b · thinks first
Fits 26.3 7 to 13 Unclear could be Moderate or Fast not measured yet Estimated
Llama 3.1 70B
llama3.1:70b
Won't fit 51.5 not known n/a not measured yet Estimated
gpt-oss 120B
gpt-oss:120b
Won't fit 70.5 not known n/a not measured yet Estimated

Models marked "thinks first" can write out their reasoning before they answer, so a job takes longer than the speed alone suggests.

With 96 GB of memory

About 67.2 GB of it can hold a model.

ModelFitNeeds (GB)Writing speed (tokens per second)SpeedQualityBasis
Llama 3.2 3B
llama3.2:3b
Fits 5 70 to 130 Fast not measured yet Estimated
Qwen3 4B
qwen3:4b · thinks first
Fits 6 56 to 104 Fast not measured yet Estimated
Gemma 3 4B
gemma3:4b
Fits 6.7 42 to 79 Fast not measured yet Estimated
Mistral 7B v0.3
mistral:7b
Fits 7.8 32 to 59 Fast not measured yet Estimated
Llama 3.1 8B
llama3.1:8b
Fits 8.3 29 to 53 Fast not measured yet Estimated
Qwen3 8B
qwen3:8b · thinks first
Fits 8.9 27 to 50 Fast not measured yet Estimated
Gemma 3 12B
gemma3:12b
Fits 15.9 17 to 32 Fast not measured yet Estimated
DeepSeek-R1 Distill Qwen 14B
deepseek-r1:14b · thinks first
Fits 13.7 16 to 29 Fast not measured yet Estimated
Phi-4 14B
phi4:14b
Fits 13.9 15 to 29 Fast not measured yet Estimated
Qwen3 14B
qwen3:14b · thinks first
Fits 13.4 15 to 28 Fast not measured yet Estimated
gpt-oss 20B
gpt-oss:20b · thinks first
Fits 16.5 35 to 65 Fast not measured yet Estimated
Gemma 3 27B
gemma3:27b
Fits 27.2 8.2 to 15 Unclear could be Moderate or Fast not measured yet Estimated
Qwen3 30B (A3B)
qwen3:30b · thinks first
Fits 22.6 41 to 76 Fast 9.3/10 Estimated
Qwen3 32B
qwen3:32b · thinks first
Fits 26.3 7 to 13 Unclear could be Moderate or Fast not measured yet Estimated
Llama 3.1 70B
llama3.1:70b
Fits 51.5 3.3 to 6 Unclear could be Slow or Moderate not measured yet Estimated
gpt-oss 120B
gpt-oss:120b
Won't fit 70.5 not known n/a not measured yet Estimated

Models marked "thinks first" can write out their reasoning before they answer, so a job takes longer than the speed alone suggests.

Reported by others

Other people ran the standard llama-bench test on this chip and published the results. They are not Findra scores. The test uses a small 7-billion-parameter model, so larger models will be slower than these figures. All published benchmarks.

SetupModel and sizeWriting speed (tokens per second)Reading speed (tokens per second)Source
30 graphics cores Llama 2 7B, Q8_0 40.0 540 llama.cpp discussion 4167
read 2026-10-10
30 graphics cores Llama 2 7B, Q4_0 61.0 538 llama.cpp discussion 4167
read 2026-10-10
38 graphics cores Llama 2 7B, Q8_0 41.8 678 llama.cpp discussion 4167
read 2026-10-10
38 graphics cores Llama 2 7B, Q4_0 66.0 671 llama.cpp discussion 4167
read 2026-10-10

Reported by others, not tested by us. Results come from a public community thread; builds and settings differ between submitters.

Rendering speed (Blender Open Data)

Blender Open Data publishes the median render score of the Cycles test for each chip, from results people submitted. Higher is faster. These are not Findra scores; the table says what each score was rendered on. Scores from different Blender versions are not directly comparable.

ChipRendered onScoreSource
Apple M2 Maxprocessor, Blender 5.2.0244Blender Open Data
read 2026-10-22

Editing, gaming, CAD and other work

Each app below publishes a minimum and a recommended system requirement. We compare them with this laptop's memory and graphics memory. We do not judge the graphics chip's class, the processor generation or the operating system, because the makers describe those in words: check the source link for the exact card, processor and system they name. Many apps and games run only on Windows or only on macOS. Integrated graphics count as below any requirement for a dedicated card. Not known means the maker does not state this laptop's graphics memory.

App32 GB64 GB96 GB
Video editing
Adobe Premiere ProRecommendedRecommendedRecommended
Adobe After EffectsRecommendedRecommendedRecommended
Final Cut ProRecommendedRecommendedRecommended
OBS StudioMinimumMinimumMinimum
Photo editing and design
Adobe PhotoshopRecommendedRecommendedRecommended
Adobe Lightroom ClassicRecommendedRecommendedRecommended
Adobe IllustratorRecommendedRecommendedRecommended
3D, CAD and rendering
BlenderRecommendedRecommendedRecommended
Autodesk AutoCADRecommendedRecommendedRecommended
SOLIDWORKSRecommendedRecommendedRecommended
Autodesk FusionRecommendedRecommendedRecommended
Autodesk RevitMinimumMinimumMinimum
SketchUpMinimumMinimumMinimum
Autodesk MayaRecommendedRecommendedRecommended
Maxon Cinema 4DRecommendedRecommendedRecommended
RhinoRecommendedRecommendedRecommended
KeyShot StudioRecommendedRecommendedRecommended
Unreal Engine 5 editorRecommendedRecommendedRecommended
Unity editorMinimumMinimumMinimum
Gaming
Cyberpunk 2077RecommendedRecommendedRecommended
Baldur's Gate 3RecommendedRecommendedRecommended
Elden RingRecommendedRecommendedRecommended
Black Myth: WukongRecommendedRecommendedRecommended
StarfieldRecommendedRecommendedRecommended
Hogwarts LegacyRecommendedRecommendedRecommended
Counter-Strike 2MinimumMinimumMinimum
FortniteRecommendedRecommendedRecommended
Call of Duty: WarzoneRecommendedRecommendedRecommended
Microsoft Flight Simulator 2024RecommendedRecommendedRecommended
Alan Wake 2RecommendedRecommendedRecommended
Forza Horizon 5RecommendedRecommendedRecommended
Red Dead Redemption 2RecommendedRecommendedRecommended
Grand Theft Auto VRecommendedRecommendedRecommended
Marvel RivalsRecommendedRecommendedRecommended
Helldivers 2RecommendedRecommendedRecommended
Hades IIRecommendedRecommendedRecommended
Stardew ValleyMinimumMinimumMinimum
Minecraft Java EditionRecommendedRecommendedRecommended
AI image and video generation
InvokeAI with SD 1.5 modelsMinimumMinimumMinimum
InvokeAI with SDXL modelsMinimumMinimumMinimum
InvokeAI with FLUX.1 modelsMinimumMinimumMinimum
Music and audio production
Ableton Live 12MinimumMinimumMinimum
FL StudioMinimumMinimumMinimum
Logic ProMinimumMinimumMinimum
Pro ToolsMinimumMinimumMinimum
Steinberg CubaseMinimumMinimumMinimum

Labels follow the makers' published lists. All use cases.

Speed is the speed of writing an answer, shown as a range. Reading speed is not estimated. How the estimates work · How close they are · Back to the finder