← Best models

Best model for a Mac with 48 GB of memory

Ranked models that need at most 28 GB, best Overall score first, then the fastest of those within 10 points of the best one that fits. The memory a model needs is an estimate: the model file size plus about 20% for the context of a single conversation.

What this page assumes. This machine shares its 48 GB with the operating system and every other program, so the page counts only 60% of it, 28 GB, as available for a model. A model is listed when its estimated memory need is at most that budget. Most models here were run with a context window of 16,384 tokens, so a longer context needs more memory than shown.

Speeds are not for this machine. Every speed on this page is tokens per second measured on the benchmark's own laptop, an AMD Ryzen AI Max+ 395 with 128 GB of unified memory. That is not a Mac with 48 GB of memory, so only the memory figures carry over; its speed will differ.

The best model that fits scores 99.3, so the fastest list covers Overall 89.3 to 99.3. 48 models are inside it. 1 of them has no single-stream speed measurement and is not listed there.

211 of 227 ranked models fit. 5 ranked models have no file size on record, so their memory need cannot be estimated and they are not listed.

Read the memory figure as an estimate, not a guarantee. The memory a model needs is estimated from its file: the weights plus about 20% for the context of a single conversation. It is not a measurement, and a long conversation or several at once needs more. A model with no file size on record is not counted as fitting any amount of memory. Speeds, and the GPU memory shown on the speed page, were measured on this benchmark's server, not on yours. That server kept 1 to 8 answers in flight at a time and loaded a context window of 16,384 tokens for most models; 42 of the models ranked ran at a shorter context. Speeds were measured on one laptop and are comparable only between models on the same backend. See the settings and the memory method.

Up to 28 GB

gemma-4-31B-it is the top-scoring ranked model that fits the 28 GB this page assumes: Overall 99.3 (95% interval 98.3–100.0), rank range 1–5, needing about 22.4 GB (file size plus 20%). It is not clearly ahead of Gemma 4 12B IT, the second: with this battery the two cannot be told apart with confidence.

Up to 28 GB: the top 5 of 211.
Place Model Overall Rank range Decode (tok/s) Memory needed (GB, est.) Size class
1 gemma-4-31B-it 99.398.3–100.0 1–5 3.8ROCm 7.1 22.4 16–35B
2 Gemma 4 12B IT 98.697.0–99.8 1–12 8.1ROCm 7.1 8.9 9–16B
3 Muse Glimmer 30B 97.996.0–99.3 1–23 3.6ROCm 7.1 21.8 16–35B
4 gemma-4-26B-A4B-it 97.895.8–99.3 1–22 15.7ROCm 7.1 21.6 16–35B
5 Qwen3.5-27B 96.293.8–98.3 3–35 2.7ROCm 7.1 19.9 16–35B

Fastest of the good ones that fit

Qwen3-VL-30B-A3B-Instruct is the fastest of the 47 models in the window: 26.9 tokens per second on our laptop (measured on Vulkan; speeds on different backends are not directly comparable), with an Overall of 92.9 (95% interval 90.1–95.5), 6.4 points below the best model that fits, gemma-4-31B-it (99.3).

Fastest of the good ones that fit: the top 5 of 47.
Place Model Overall Rank range Decode (tok/s) Memory needed (GB, est.) Size class
1 Qwen3-VL-30B-A3B-Instruct 92.990.1–95.5 7–65 26.9Vulkan 23.5 16–35B
2 Qwen3.5-35B-A3B 94.892.1–97.1 3–60 23.2Vulkan 25.4 16–35B
3 Laguna XS 2.1 91.287.7–94.4 6–73 23.1ROCm 7.1 24.3 —
4 Sage-A1.5-Q4_K_M 94.591.8–96.7 5–56 22.6ROCm 7.1 26.4 > 35B
5 Qwen2.5-14B-Instruct 91.388.0–94.2 11–72 21.7ROCm 7.1 10.8 9–16B

A higher place is not always a clear win. When two models have overlapping rank ranges, this battery cannot tell them apart; see how sure a score is. A dash means not measured or not scored, which is different from 0. Speeds were measured on one laptop and are comparable only between models on the same backend.

More: all the best-model pages, the hardware picker, speed and memory for every model.