MacBook Pro M5 Pro and M5 Max for local AI: 70B models on the move (2026)

By /
MacBook Pro with M5 Pro and M5 Max on Apple UK, October 2026

The MacBook Pro with M5 Pro or M5 Max is Apple's laptop for local AI, with up to 128 GB. Buy the 14-inch M5 Max with the 40-core GPU and 128 GB: £6,699 at Apple, plus a power adapter, which UK models lack. One published test ran Llama 3.3 70B on it at about 10 tokens a second. The M5 Pro starts at £2,499. I rate it 4 out of 5. If it will stay on a desk, see which Mac or mini PC to buy for local AI.

Last updated 22 October 2026. Written by Annie, a search and digital marketing specialist. Based on the maker's specifications, UK prices and published benchmarks; I have not tested this machine myself. See the editorial policy.

MacBook Pro M5 Pro and M5 Max at a glance for local AI: £2,499 for the M5 Pro with 24 GB, up to 128 GB of memory, runs 70B models on 64 GB and up, rated 4 out of 5
MacBook Pro M5 Pro and M5 Max at a glance. Prices from UK stores, checked 17 October 2026.

Key facts

  • What it is: Apple's 14-inch and 16-inch pro laptop. The M5 Pro and M5 Max versions were announced on 3 March 2026 and went on sale on 11 March 2026. The 14-inch also comes with the plain M5 chip (up to 32 GB, 153 GB/s), which this page does not cover.
  • Chip: M5 Pro with a 15-core CPU and 16-core GPU, or an 18-core CPU and 20-core GPU. M5 Max with an 18-core CPU (6 super cores, 12 performance cores) and a 32-core or 40-core GPU. Every GPU core has a Neural Accelerator.
  • Memory options: M5 Pro: 24 GB, 48 GB or 64 GB (64 GB needs the 20-core GPU). M5 Max: 36 GB with the 32-core GPU, or 48 GB, 64 GB or 128 GB with the 40-core GPU. You choose at purchase. It cannot be upgraded later.
  • Usable for AI: about 16 GB (24 GB model), 27 GB (36 GB), 36 GB (48 GB), 48 GB (64 GB) or 96 GB (128 GB) can go to the GPU by default. You can raise it with one Terminal command (see Software below).
  • Memory bandwidth: 307 GB/s (M5 Pro), 460 GB/s (M5 Max, 32-core GPU), 614 GB/s (M5 Max, 40-core GPU)
  • UK price: from £2,499 (14-inch M5 Pro, 24 GB, 1 TB), or £6,699 for 128 GB (14-inch M5 Max, 40-core GPU, 2 TB, built to order), plus £59-£99 for a power adapter, as checked 17 October 2026.
  • Where to buy: Apple UK for every memory option; John Lewis for the standard models; Apple's refurbished store for some M5 Max models.
  • Best for: running 32B to 70B models away from a desk, and developers who want one machine for everything.
  • Our rating (desk review, not hands-on): ★★★★☆ 4/5

On this page

Which AI models can the MacBook Pro M5 run?

On these memory sizes macOS gives the GPU about three quarters of the memory (two thirds on the 24 GB model), so plan around the figures in brackets.

Memory optionBiggest model at 4-bitExample modelsPublished speed
M5 Pro, 24 GB (about 16 GB for the GPU)14B with long chats (Qwen3 14B is 8.5 GB)Llama 3.1 8B (4.8 GB), Qwen3 14BLlama 2 7B, 4-bit (Q4_0), llama.cpp: about 66-68 tokens a second (llama.cpp Apple silicon thread)
M5 Max 32-core GPU, 36 GB (about 27 GB)32B (19.4 GB) with long chatsQwen3.8 27B, 32B modelsNo published figure for the 32-core M5 Max
M5 Pro or M5 Max, 48 GB (about 36 GB)32B with long chats; Qwen3 30B-A3B at 8-bit (about 32 GB)Qwen3.8 27B, 32B modelsNo published figure
M5 Pro or M5 Max, 64 GB (about 48 GB)70B (Llama 3.3 70B is 42.4 GB) with a normal-length chatLlama 3.3 70BNo published 70B figure at 64 GB
M5 Max 40-core GPU, 128 GB (about 96 GB)70B at 8-bit (about 74 GB); 80B mixture of experts at 8-bitLlama 3.3 70B, Qwen3-Next 80BLlama 3.3 70B, 4-bit (Q4_K_M), LM Studio: about 10 tokens a second (ifanr); Llama 2 7B, Q4_0, llama.cpp: about 120

A long conversation adds 2-4 GB on top of the model file. Mixture-of-experts models such as Qwen3 30B-A3B or Qwen3-Next 80B read only a small part of their weights for each word, so they answer much faster than a dense 70B model, but they still need memory for all of it. The 235B Qwen3 model (142.5 GB at 4-bit) does not fit even in 128 GB.

How fast is it?

Two sets of published figures stand out: the llama.cpp project's Apple silicon benchmark thread, and a hands-on 70B test by Chinese tech site ifanr.

  • M5 Max, 40-core GPU, Llama 2 7B, 4-bit (Q4_0), llama.cpp: about 120 tokens a second when writing and about 3,220 tokens a second when reading the prompt. At 8-bit (Q8_0) it wrote about 72 tokens a second.
  • M5 Pro, same test: about 68 tokens a second (16-core GPU) and 66 (20-core GPU), with prompt reading of about 1,340 and 1,621.
  • Against the M4 Max: the 40-core M4 Max wrote about 83 tokens a second and read prompts at about 886 in the same thread. The 40-core M5 Max writes about 44% faster and reads prompts about 3.6 times faster.
  • M5 Max, 128 GB, Llama 3.3 70B, 4-bit (Q4_K_M), LM Studio: about 10 tokens a second, measured by ifanr on a 14-inch test machine. The 8-bit version would not load at the 128k context setting they tried.
  • Apple's claim: up to 4 times faster LLM prompt processing than the M4 Max, and up to 3.9 times for the M5 Pro against the M4 Pro. Apple did not name the model.
  • No published figures for the 32-core M5 Max, or for 70B models on the M5 Pro, by 17 October 2026.

The Mac mini with M5 Pro uses the same M5 Pro chip, so expect the same speeds from it. The Mac mini M6 wrote about 36 tokens a second on the 7B test.

Which version should you buy?

Buy the 14-inch M5 Max with the 40-core GPU and 128 GB, at £6,699, and add the 96 W adapter for £79. It holds 70B models at 8-bit and has 614 GB/s memory, twice the M5 Pro's. Choose the 16-inch (£6,999, and it needs the £99 140 W adapter) only if you want the bigger screen and the 100 Wh battery.

If that is too much, buy the 14-inch M5 Pro with the 20-core GPU and 64 GB, at £3,699. It holds Llama 3.3 70B at 4-bit, but nobody has published its speed, and with half the bandwidth it will be slow at that size. Skip the M5 Max with 36 GB, which costs £4,099 for about 27 GB of usable memory. Skip the 24 GB models for AI.

UK prices and where to buy

VersionMemoryStorageUK priceWhere to buy
14-inch M5 Pro, 15-core CPU, 16-core GPU24 GB1 TB£2,499Apple UK, John Lewis
14-inch M5 Pro, 15-core CPU, 16-core GPU, built to order48 GB1 TB£3,099Apple UK
14-inch M5 Pro, 18-core CPU, 20-core GPU, built to order64 GB1 TB£3,699Apple UK
14-inch M5 Max, 32-core GPU36 GB2 TB£4,099Apple UK, John Lewis
14-inch M5 Max, 40-core GPU, built to order64 GB2 TB£5,099Apple UK
14-inch M5 Max, 40-core GPU, built to order (our pick)128 GB2 TB£6,699Apple UK
16-inch M5 Pro, 20-core GPU48 GB1 TB£3,599Apple UK, John Lewis
16-inch M5 Pro, 20-core GPU, built to order64 GB1 TB£3,999Apple UK
16-inch M5 Max, 40-core GPU48 GB2 TB£4,999Apple UK, John Lewis
16-inch M5 Max, 40-core GPU, built to order128 GB2 TB£6,999Apple UK

Apple's UK refurbished store had the 16-inch M5 Max with 40-core GPU, 48 GB and 2 TB at £4,249, £750 below new. Students and staff pay £6,139 for our 128 GB pick and £3,449 for the 64 GB M5 Pro in Apple's UK education store.

Hidden costs

  • Power adapter. UK models come with only the MagSafe cable. Apple sells the 70 W adapter for £59, the 96 W for £79 and the 140 W for £99. The 16-inch needs at least 140 W to charge, and Apple says the 96 W and 140 W adapters support High Power Mode.
  • Memory you can't add later. On the 14-inch M5 Max with 40-core GPU, going from 48 GB to 128 GB costs £2,000 at purchase.
  • Storage. M5 Pro models start at 1 TB and M5 Max at 2 TB. Moving the 14-inch M5 Pro (20-core GPU, 24 GB) from 1 TB to 2 TB adds £500.
  • AppleCare+. £99.99 a year or £9.99 a month for the 14-inch, and £149.99 a year or £14.99 a month for the 16-inch.

Software: Ollama, LM Studio and the rest

Everything runs natively on Apple silicon, on battery or plugged in, with no drivers to install.

  • MLX: Apple's own framework. Apple says MLX uses the M5 family's GPU Neural Accelerators from macOS 26.2, and its tests on the base M5 showed up to 4 times faster time to first token than the M4, with 19-27% faster writing.
  • Ollama: runs models through llama.cpp with Metal, and since a March 2026 preview through MLX as well. Ollama says it uses the Neural Accelerators on M5, M5 Pro and M5 Max chips to speed up both first response and writing. The MLX preview asked for more than 32 GB of memory. See our Ollama guide.
  • LM Studio: uses both MLX and llama.cpp and needs macOS 14 or later. ifanr ran its 70B test in LM Studio. See our LM Studio guide.
  • llama.cpp: runs on the GPU through Metal.
  • One setup step: to give the GPU more room on a 128 GB machine, run sudo sysctl iogpu.wired_limit_mb=114688 in Terminal (that sets 112 GB). It resets when you restart unless you add it to /etc/sysctl.conf. Leave several gigabytes for macOS.

Noise, power and size

  • 14-inch: 1.55 x 31.26 x 22.12 cm, 1.60 kg (M5 Pro) or 1.62 kg (M5 Max), 72.4 Wh battery.
  • 16-inch: 1.68 x 35.57 x 24.81 cm, 2.14 kg (M5 Pro) or 2.15 kg (M5 Max), 100 Wh battery.
  • Noise and power (16-inch M5 Max, Notebookcheck): maximum fan noise of 24.8 dB(A) in Low Power mode, 40.9 dB(A) in Automatic and 53.9 dB(A) in High Power. The CPU and GPU together peaked at 114 W, then settled at about 70 W.
  • Battery under load: Notebookcheck found the battery still dropped about 12% an hour at full load while plugged into the 140 W adapter. Results on battery matched results on mains power.

Who makes the MacBook Pro M5?

  • Apple Inc., based at One Apple Park Way, Cupertino, California. Apple introduced the Macintosh in 1984.
  • UK support comes through Apple Stores, the online store and 0800 048 0408.
  • Apple gives a one-year limited warranty. Under English law you also have up to six years to claim from the seller for faulty goods.

What's good

  • Up to 128 GB, about 96 GB of it for the GPU: room for Llama 3.3 70B at 8-bit.
  • 614 GB/s on the 40-core M5 Max: about 120 tokens a second on the llama.cpp 7B test and 3.6 times the M4 Max's prompt speed.
  • A published 70B result: about 10 tokens a second at 4-bit in LM Studio.
  • Full speed on battery, according to Notebookcheck.
  • Easy to buy in the UK, with John Lewis guarantees of 2-3 years on standard models and refurbished M5 Max models from Apple.

What's not so good

  • 128 GB costs £6,699 before the adapter, about £70 for each gigabyte the GPU can use.
  • No power adapter in the UK box.
  • Notebookcheck found the 140 W adapter cannot keep up at full load, and the 16-inch cooling holds the M5 Max at about 70 W.
  • The memory options that matter for AI are built to order, so only Apple sells them. 64 GB on the M5 Pro needs the 20-core GPU, and 128 GB needs the 40-core M5 Max.
  • UK prices rose by £300-£500 in June 2026.

Who should buy it?

Buy it if you need 32B to 70B models away from a desk: consultants working on confidential documents, developers who travel, or anyone who wants one machine for everything. Skip it if the machine will live on a desk. The Mac mini with M5 Pro holds a 70B model with 64 GB from £2,699, and the Mac mini M6 handles 8B-27B models for £1,299.

Buying in the UK

  • Availability: Apple UK sells every version. John Lewis stocks the standard models at Apple's prices. Prices include 20% VAT and there are no import charges.
  • Warranty: Apple gives one year. John Lewis includes a 2-year guarantee on the models I checked, and 3 years on the 14-inch M5 Max with 36 GB.
  • Returns: Apple accepts returns within 14 days of delivery. Refurbished models carry the same one-year warranty and 14-day returns.
  • 2026 prices: Apple raised UK MacBook Pro prices on 25 June 2026, blaming memory shortages. The 14-inch M5 Max launched at £3,599 in March and now costs £4,099.

Interesting facts about the MacBook Pro M5

  • UK MacBook Pros ship without a power adapter, and the 16-inch needs at least 140 W to charge.
  • The M5 Pro and M5 Max use what Apple calls Fusion Architecture, joining two dies into one chip. Apple says the SSD reaches up to 14.5 GB/s.
  • Ollama demonstrated its June 2026 MLX engine update on a MacBook Pro with M5 Max.

Alternatives to the MacBook Pro M5

  • Mac mini with M5 Pro: pick it for the same M5 Pro chip and 64 GB from £2,699 if you work at a desk.
  • Mac Studio M5 Max: pick it for the M5 Max chip on a desk, with no battery or separate power adapter to think about.
  • Mac mini M6: pick it if 8B-27B models are enough and you want to spend £1,299.

Our rating

AreaScore
Memory for AI4/5
Speed4/5
Value2/5
Software5/5
UK buying5/5
Overall★★★★☆ 4/5

The M5 Max with 128 GB is the most capable laptop Apple sells for local AI, and every Mac AI tool works on it. You pay heavily for the memory, so buy it only if the models need to travel with you.

Frequently asked questions

Can the MacBook Pro M5 Max run Llama 70B?

Yes. With 64 GB it holds Llama 3.3 70B at 4-bit (42.4 GB), and with 128 GB the 8-bit version too. ifanr measured about 10 tokens a second at 4-bit on a 128 GB M5 Max in LM Studio.

How much memory does a MacBook Pro need for local AI?

64 GB for 70B models at 4-bit, and 128 GB for 70B at 8-bit or large mixture-of-experts models. 48 GB covers 32B models with long chats.

Should I buy the M5 Pro or M5 Max for local AI?

The 40-core M5 Max if you can afford it. It has twice the memory bandwidth (614 GB/s against 307 GB/s), wrote about 120 tokens a second against about 66 on the same llama.cpp test, and is the only version with 128 GB.

Sources

More guides

Scroll to Top