Apple M5 Pro

Apple M5 Pro (64GB): the maximum-memory M5 Pro configuration, announced March 3, 2026 for the 14-inch and 16-inch MacBook Pro on the full 18-core CPU (6 super cores + 12 performance cores)/20-core GPU die at 307 GB/s — the same bandwidth as the 48GB build, since Apple doesn't cut bandwidth for the smaller M5 Pro die. It's also a new tier: M4 Pro topped out at 48GB, so 64GB is this generation's first Pro-tier memory increase.

With 56 GB of real headroom, this is the first M5 Pro configuration where large MoE models really stretch out: Qwen 3.5 122B-A10B fits at Q2_K (54.47 GB, 16.5 tok/s) and Nemotron 3 Super 120B reaches Q2_K (52.03 GB, 15.4 tok/s) — both too large for any smaller M5 Pro tier. Dense 70B-class models clear a real quantization too: Llama 3.3 70B and Qwen2.5 72B both reach Q4_K_M (50.75 GB and 52.12 GB) at 5.4 and 5.3 tok/s. 56 of the 84 tracked models fit natively — the same count as the M5 Max 64GB, at exactly half the speed on every shared fit, since the M5 Max 64GB's 614 GB/s is exactly double this configuration's 307 GB/s.

MLX and llama.cpp's Metal backend are both fully supported. At the same capacity as the M5 Max 64GB but half the bandwidth, this configuration reads as the capacity-first choice within the M5 Pro line — the M5 Max 64GB fits the identical model list at 2x the decode speed, for a real price premium.

Apple M5 Pro (48GB): the upgraded M5 Pro configuration, available in the 14-inch and 16-inch MacBook Pro, standard on the full 18-core CPU (6 super cores + 12 performance cores)/20-core GPU die, at 307 GB/s — a 12.5% jump over the M4 Pro generation's 273 GB/s and the second-largest single-generation Pro-tier bandwidth increase Apple has shipped, after M4 Pro's own recovery (see this page's bandwidth note for the full multi-generation picture).

With 40 GB of real headroom, this calculator's highest-fitting quant for a dense 70B model here is an aggressive Q2_K (32.88 GB, 8.4 tok/s) — technically native, but a real quality trade-off, and about 13.5% faster than the identically-sized M4 Pro 48GB's 7.4 tok/s on the same fit. MoE models fare far better at this size: Qwen 3.5 35B-A3B reaches near-full Q6_K (32.37 GB, 29.3 tok/s), and Command-R 35B fits at Q5_K_M (39.94 GB, 6.9 tok/s). 51 of the 84 tracked models fit natively.

Full llama.cpp K-quants support and MLX optimization. This configuration fits the identical 51-model list as the M5 Max 48GB but at exactly half the decode speed (307 vs 614 GB/s) — the Max tier's premium here buys bandwidth, not extra model compatibility.

Apple M5 Pro (24GB): the base M5 Pro configuration. On the 14-inch MacBook Pro it's standard on a cut-down 15-core CPU (5 super cores + 10 performance cores)/16-core GPU die; the 16-inch MacBook Pro's base M5 Pro instead ships the full 18-core CPU/20-core GPU die at this same 24GB — but both bins share the identical 307 GB/s memory bandwidth, since Apple doesn't cut bandwidth for the smaller die on this chip.

With 16 GB of real headroom, this configuration is where mixture-of-experts models pull dramatically ahead of same-size dense ones: Qwen 3.5 35B-A3B — a 35B-parameter MoE model — fits at Q2_K (15.12 GB) at 61.7 tok/s, essentially real-time, while the dense Qwen3 32B needs the same Q2_K precision at a near-identical footprint (15.50 GB) but manages only 17.7 tok/s, a 3.5x gap from MoE decode reading far fewer bytes per token. Llama 3.1 8B and Qwen2.5 7B both clear Q8_0, near-full precision, at 25.6 and 28.7 tok/s — about 12.5% faster than the identically-sized M4 Pro 24GB on the same two fits (22.8 and 25.5 tok/s there). 39 of the 84 tracked models fit natively.

MLX and llama.cpp's Metal backend are both mature here. Nothing this configuration fits comes close to the default ~18 GB Metal working-set limit (75% of 24 GB) — the largest fit is around 15.5 GB — so the manual wired_limit override doesn't apply at this capacity.