Best Laptops for AI and ML Developers (2026)

TL;DR

• Best overall: Apple MacBook Pro 16 (M5 Max, 128GB) — Check price

There are two ways to run AI models on a laptop in 2026, and StorageReview's lab, which ranks laptops by measured local-AI performance, describes both: discrete NVIDIA GPUs give the most tokens per second, but the model has to fit in the card's own 8GB to 24GB of memory; unified-memory designs let the GPU borrow most of system memory, so far larger models load at all, at lower speed. [src2] The picks follow that split. The Apple MacBook Pro 16 with the M5 Max and 128GB is the capacity pick with the most bandwidth (614GB/s); the ASUS ROG Flow Z13 with 128GB is the Windows capacity pick; the ASUS ROG Strix Scar 18 and the Dell Pro Max 16 Plus are the CUDA picks, for models that fit in their 24GB. [src7, src1, src17, src2]

The MacBook keeps the top spot on the reviews. Notebookcheck scores the 16-inch M5 Max 95% for AI performance and found its GPU within 12% of the 120W RTX 5090 in the ASUS ProArt P16 in 3DMark Steel Nomad; Macworld calls it "Apple's fastest portable workstation", and Engadget "the professional notebook to beat". [src3, src4, src5] Hardware Corner's look at the M5 Max with 128GB found a 122-billion-parameter model generating about 55 to 65 tokens per second, but also that a desktop RTX 5090 generated tokens two to three times faster on a similar-sized dense model that fit in its 32GB. [src6] The 128GB Amazon listing we link had no purchasable new offer in our last three daily checks, through 2026-10-09; Apple's US store still offered the 128GB option, built to order, on 2026-10-10. [src28]

Two things changed in this refresh. The workstation pick moved from the Lenovo ThinkPad P16 Gen 3 to the Dell Pro Max 16 Plus, because both labs that measured the two put the Dell ahead: StorageReview at 178.6 against 141.2 tokens per second on Phi with the same RTX PRO 5000 GPU, and Notebookcheck, whose P16 Gen 3 had an RTX PRO 3000 limited to 105W, at about 24% in professional and 3DMark tests. [src2, src14] The P16 Gen 3 stays on the card for buyers who need ECC memory, which the Dell does not offer. [src16, src11] Dell's US store no longer listed the RTX PRO 5000 build of the 16 Plus when we checked on 2026-10-10, while the larger 18 Plus is still offered with that GPU. [src30, src29] And the AMD Ryzen AI Max+ machines arrived on the evidence: Tom's Guide's best AI laptop overall is the Flow Z13, and StorageReview's best laptop for large models is HP's ZBook Ultra G1a, built on the same AMD platform. [src1, src2] The Microsoft Surface Laptop 7 is gone: no source we found tested it for AI development, and StorageReview's advice is to buy for the GPU and memory, not the NPU. [src2]

Best Overall (Largest Local Models): Apple MacBook Pro 16 (M5 Max, 128GB) — Check price

Apple's 40-core M5 Max pairs an 18-core CPU with 614GB/s of memory bandwidth and up to 128GB of unified memory in a 4.7 lb laptop with a 100Wh battery. [src7] Apple says the M5 Max processes LLM prompts up to 4x faster than the M4 Max. [src8] Macworld tested the 128GB configuration (with 4TB, 6,149 USD as tested) and calls it "Apple's fastest portable workstation", with a 32% jump in GPU performance over the M4 Max in Cinebench 2024 and 24 hours 35 minutes in its video-loop battery test. [src4] Notebookcheck rates the 16-inch M5 Max 95% for AI performance. It notes that the M5 Max GPU "can access the system memory, which means it has an advantage over most Windows competitors with Nvidia GPUs", and found it only 12% slower in 3DMark Steel Nomad than the RTX 5090 in the ASUS ProArt P16, which runs at 120W. Its criticisms are the cooling, which holds the chip to about 70W under sustained load, and the 140W adapter: the battery fell 12% in an hour of full load, even on a 180W charger. It measured 18.5 hours of Wi-Fi browsing at 150 nits. [src3] Hardware Corner's look at a 14-inch M5 Max with the same 128GB found a 122-billion-parameter mixture-of-experts model generating about 55 to 65 tokens per second while using 72GB to 76GB, which it calls "the ability to run 120B class models locally on a single portable system". [src6] Engadget calls the M5 Max MacBook Pro "once again the professional notebook to beat". [src5] The 128GB Amazon listing we link had no purchasable new offer in our last three daily checks, through 2026-10-09. When we checked on 2026-10-10, Apple's US store still offered 128GB of unified memory on the 40-core M5 Max, built to order. [src28]

Best CUDA Training: ASUS ROG Strix Scar 18 (2025, RTX 5090) — Check price

Tom's Guide makes the Strix Scar 18 its best AI laptop for power users. With the Core Ultra 9 275HX and RTX 5090 it counts "1,837 TOPS" of AI processing and says the machine could "even train my own AI models"; its one drawback is the steep price. [src1] Notebookcheck measured the RTX 5090 Laptop GPU, with 24GB of VRAM, at the full 175W (150W plus 25W Dynamic Boost) and found its performance constant under sustained load. It praises access to the components without a screwdriver; against the laptop are loud fans in Performance and Turbo modes, high power draw, a 3.3 kg (7.3 lb) weight, and a GPU limited to 55W on battery, where its performance fell 64%. [src9] Ultrabook Review's G835LX cost 4,299 USD when reviewed; it found the RTX 5090 adds roughly 10% to 15% GPU performance over the RTX 5080 version, for a 25% higher price. [src10] For AI work, the RTX 5090's 24GB of VRAM is the reason to pay that premium: no discrete GPU on this card has more. [src9]

Best Workstation: Dell Pro Max 16 Plus (RTX PRO 5000) — Check price

StorageReview's lab measured the Dell Pro Max 16 Plus with the RTX PRO 5000 Blackwell (24GB) at 178.6 tokens per second on Phi, 134.2 on Mistral and 114.7 on Llama 3 in UL Procyon's text-generation test, about 96% of the 18-inch flagship, while starting at 5.63 lb and lasting 6 hours 21 minutes in its battery test. "For most people who want serious local inference in a bag, this is the better buy of the two," it says, and it holds StorageReview's Best Overall Mobile Workstation spot. [src2] Notebookcheck tested the top configuration (Core Ultra 9 285HX, RTX PRO 5000, 128GB of CAMM2 memory, 4K 120Hz OLED; about 5,000 USD at the time) and calls it "one of the fastest and most practical workstations available". Its reservations: CPU temperatures of 95°C to 105°C under stress, a GPU that settles around 125W although rated for 175W, limited battery life, and no ECC memory option. [src11] TechRadar names it the best mobile workstation overall, for "desktop-class performance" in a laptop that still fits in a backpack. [src12] Our link goes to Dell's US product page. When we checked it on 2026-10-10, Dell listed the 16 Plus in two configurations only: a Core Ultra 7 265HX with an RTX PRO 2000 (8GB) at 5,698 USD, and a Core Ultra 9 285HX with an RTX PRO 3000 at 8,121 USD. The RTX PRO 5000 build reviewed here was not offered, and none of these reviews measured the lower GPUs in this laptop. [src30] Best you can buy today: the Dell Pro Max 18 Plus below, which Dell still configures with the RTX PRO 5000. [src29]

Best Windows Laptop for Large Local Models: ASUS ROG Flow Z13 (2025, 128GB) — Check price

Tom's Guide's best AI laptop overall. Its Ryzen AI Max+ 395 is "capable of running a 120-billion-parameter AI model in a small, portable frame", the tablet weighs 2.7 lb, and the 128GB configuration you need for the largest models costs over 3,000 USD; Tom's Guide's complaint is the sparse port selection. [src1] Windows Central calls it "one of the most capable PCs in the world for AI" and notes that up to 96GB of its memory can be assigned to the GPU. It also found that the tablet gets hot and loud, with some thermal throttling, and lasts 3 to 8 hours on battery. [src17] Ultrabook Review notes its 256-bit LPDDR5X-8000 memory bus and a weight of 1.63 kg (3.6 lb) with the keyboard folio, plus a 200W charger. [src18]

Lower-Cost M5 Max: Apple MacBook Pro 16 (M5 Max, 64GB) — Check price

The same 40-core M5 Max with 614GB/s of bandwidth, with 64GB of unified memory instead of 128GB. [src7] Notebookcheck's ratings above come from a 48GB unit. [src3] Half the memory halves the largest model you can load: Hardware Corner's 122-billion-parameter test used 72GB to 76GB, more than this configuration has. [src6]

Fastest Local Inference: Dell Pro Max 18 Plus (RTX PRO 5000) — Check price

StorageReview's fastest laptop for local AI: 185.1 tokens per second on Phi, 140.5 on Mistral and 119.7 on Llama 3, with the first token in under 0.35 seconds. It weighs 7.17 lb and ran 3 hours 39 minutes in StorageReview's battery test, the shortest in its dataset. "If local inference speed is the whole question, this is the answer." It listed at 9,245 USD as tested. [src2] TechRadar tested the same 9,245 USD configuration (Core Ultra 9, 128GB, RTX PRO 5000 24GB, 2TB, 18-inch 2560x1600) and calls it "the developer's dream machine, the coder's crutch, and the engineer's engine". It barely fits a backpack, and TechRadar says the base model starts at around 3,500 USD. [src13] When we checked Dell's US store on 2026-10-10, the 18 Plus started at 4,190 USD and offered the RTX PRO 5000 (24GB) as a 3,560 USD option. [src29]

ECC Memory Option: Lenovo ThinkPad P16 Gen 3 (RTX PRO 4000, 128GB) — Check price

Lenovo's specifications list RTX PRO 3000 (12GB), RTX PRO 4000 (16GB) and RTX PRO 5000 (24GB) Blackwell GPUs for the P16 Gen 3, all with ECC graphics memory, and system memory up to 192GB, or up to 128GB of ECC memory. [src16] That ECC option is what the Dell Pro Max 16 Plus lacks. [src11] TechRadar gave an RTX PRO 5000 unit (32GB, 3.2K tandem OLED; 6,998.95 USD) five stars and calls it "an incredible mobile workstation for the business professional"; it weighs 5.6 lb and charges over USB-C. [src15] StorageReview measured an RTX PRO 5000 unit at 141.2 tokens per second on Phi, behind both Dell Pro Max laptops with the same GPU; its unit had only 32GB of RAM. [src2] Notebookcheck says "For CAD or local AI usage, this is a great laptop", but found Lenovo limits the GPU to 105W, leaving its unit about 24% behind the Dell Pro Max 16 Plus. [src14] The configuration we link, with an RTX PRO 4000 (16GB) and 128GB, is not one these outlets tested, and it had no purchasable new offer when we checked on 2026-10-09.

RTX PRO 3000 Configuration: Lenovo ThinkPad P16 Gen 3 (RTX PRO 3000, 64GB) — Check price

Notebookcheck reviewed the P16 Gen 3 with this GPU, the RTX PRO 3000 (12GB), alongside a Core Ultra 9 285HX and 96GB, and says "For CAD or local AI usage, this is a great laptop", praising the room for 192GB of RAM and three SSDs. But Lenovo limits the GPU to 105W, which left it about 24% behind the Dell Pro Max 16 Plus in Notebookcheck's professional and 3DMark tests, and its 180W charger can fall behind under full load. Its unit weighed 2.74 kg (6.0 lb). [src14] The configuration we link pairs the same GPU with a Core Ultra 7 255HX and 64GB; ECC system memory depends on the configuration, per Lenovo's specifications. [src16]

Premium 18-Inch CUDA: Razer Blade 18 (2025, RTX 5090) — Check price

Windows Central's unit (Core Ultra 9 275HX, RTX 5090, 64GB; 5,199.99 USD list price) weighed 3.10 kg (7.06 lb) and lasted about 1 hour 25 minutes of gaming on battery. Its fans are "way too loud under stress", and Windows Central says that anything above the RTX 5070 Ti brings diminishing returns "unless you're planning to utilize local AI on the extra GDDR7 VRAM". It has Thunderbolt 5 and a dual-mode UHD+ 240Hz / FHD+ 440Hz screen. [src19] The 32GB configuration we link had no purchasable new offer when we checked on 2026-10-09.

Thinnest RTX 5090: Razer Blade 16 (2025, RTX 5090) — Check price

Tom's Hardware's unit (4,499.99 USD as tested) fits the RTX 5090 Laptop GPU (24GB GDDR7) into a chassis 0.69 inches thick and 4.72 lb, but at 160W (135W plus 25W Dynamic Boost), next to a 28W Ryzen AI 9 HX 370 that costs performance in CPU-bound games. It lasted 7 hours 21 minutes in Tom's Hardware's battery test, and the RAM is soldered. [src20] It had no purchasable new Amazon offer when we checked on 2026-10-09.

Desktop Replacement: MSI Titan 18 HX AI (2025, A2XWJG) — Check price

PC Guide's A2XWJG (Core Ultra 9 285HX, RTX 5090, 64GB) delivers "godly power, at an ungodly price"; it weighs 3.6 kg and lasted 4 hours 15 minutes in PC Guide's economy-mode battery test. [src21] KitGuru gives its unit (64GB, 6TB in RAID 0) 9 out of 10 but notes slight thermal throttling in the highest performance mode and mediocre battery life. [src22] The Amazon listing we link is the reviewed 2025 A2XWJG, in MSI's 96GB, 6TB configuration (A2XWJG-016US); it had no purchasable offer when Amazon was checked on 2026-10-10. MSI also sells a newer Titan 18 HX AI A2WJ with a Core Ultra 9 290HX, which we have not seen reviewed, so check the model number on any other listing.

Lightest RTX 5090: ASUS ProArt P16 (H7606WX, RTX 5090) — Check price

Ultrabook Review's late-2025 ProArt P16 (Ryzen AI 9 HX 370, RTX 5090 24GB, 64GB of soldered LPDDR5X, 4K 120Hz tandem OLED) weighs 1.93 kg (4.25 lb) and cost 3,999 USD at launch. Its RTX 5090 runs at about 120W, up to 130W with Dynamic Boost, and Ultrabook Review counts it among the better options if you need solid performance in a notebook under 2 kg. [src24] Notebookcheck measured the MacBook Pro 16's M5 Max GPU only 12% behind this laptop's RTX 5090 in 3DMark Steel Nomad. [src3]

18-Inch RTX 5090 Alternative: MSI Raider 18 HX AI (A2XWJG, RTX 5090) — Check price

PCWorld tested the Raider 18 HX AI with an RTX 5080 at 175W, a Core Ultra 9 285HX and two 2TB PCIe 5.0 SSDs in RAID 0 (4,499 USD) and calls it "a benchmark-breaking beast". It weighs 7.94 lb and lasted a little under four hours in PCWorld's video test. MSI also sold an RTX 5090 configuration, which PCWorld listed at 4,899 USD; that is the version we link, and PCWorld did not test it. [src23]

16-Inch OLED Alternative: Lenovo Legion Pro 7i Gen 10 (RTX 5090) — Check price

Notebookcheck's Gen 10 review unit had an RTX 5080 at 175W. It calls the laptop "a powerhouse option" for buyers who accept its size, its 400W adapter and very high core temperatures when gaming. [src25] Tom's Hardware, also testing an RTX 5080 unit (3,204 USD as tested), calls it "an outstanding 16-inch gaming laptop" with an excellent OLED screen; it weighs 6 lb and lasted 4 hours 37 minutes on battery, the shortest in its comparison. [src26] The configuration we link has the RTX 5090 and 64GB; neither review measured it.

Head-to-Head Comparisons

Apple MacBook Pro 16 (M5 Max, 128GB) vs ASUS ROG Strix Scar 18 (2025, RTX 5090)

Capacity against CUDA speed. The M5 Max's GPU can use up to 128GB of unified memory at 614GB/s; the Strix Scar 18's RTX 5090 Laptop GPU has 24GB of its own at 175W. [src7, src9] Hardware Corner found that a desktop RTX 5090 (32GB, more than the laptop version's 24GB) processed prompts almost 4x faster and generated tokens 2x to 3x faster than the M5 Max on a similar-sized dense model that fit in the desktop card's memory, while the M5 Max handled a 64K context the RTX 5090 could not load and ran 120B-class models that do not fit on an RTX 5090 at all. [src6] Notebookcheck measured the M5 Max GPU within 12% of an RTX 5090 running at 120W. [src3] Away from the wall the gap widens: the Strix Scar 18's GPU drops to 55W on battery, while Notebookcheck measured 18.5 hours of browsing from the MacBook. [src9, src3]

Pick the Apple MacBook Pro 16 (M5 Max, 128GB) if: your models are larger than 24GB, you work in MLX or llama.cpp on macOS, or you need to work away from a socket.
Pick the ASUS ROG Strix Scar 18 (2025, RTX 5090) if: your tools need CUDA, your models fit in 24GB, and the laptop will stay plugged in.

Dell Pro Max 16 Plus (RTX PRO 5000) vs Lenovo ThinkPad P16 Gen 3

Both are 16-inch workstations with RTX PRO Blackwell options. StorageReview measured both with the RTX PRO 5000: 178.6 tokens per second on Phi for the Dell against 141.2 for the Lenovo, whose unit had only 32GB of RAM. [src2] Notebookcheck found the P16 Gen 3's RTX PRO 3000, limited to 105W, about 24% behind the Dell in professional and 3DMark tests. [src14] The Lenovo's advantages are ECC system memory (up to 128GB) and room for 192GB of non-ECC RAM and three SSDs; Notebookcheck lists no ECC option for the Dell. [src16, src14, src11]

Pick the Dell Pro Max 16 Plus if: local-AI throughput is what you are buying.
Pick the ThinkPad P16 Gen 3 if: your organisation requires ECC system memory, or you need more than 128GB of RAM.

ASUS ROG Flow Z13 (2025, 128GB) vs Apple MacBook Pro 16 (M5 Max, 128GB)

Both put large models within reach of the GPU. The Flow Z13 can assign up to 96GB to its integrated GPU over a 256-bit LPDDR5X-8000 bus; the M5 Max's GPU reaches all 128GB at Apple's stated 614GB/s. [src17, src18, src7] Tom's Guide says the Flow Z13 can run a 120-billion-parameter model, and Hardware Corner measured a 122-billion-parameter model on the 128GB M5 Max at about 55 to 65 tokens per second. [src1, src6] The Flow Z13 is a 2.7 lb Windows tablet that Windows Central found hot and loud under load; the MacBook is the more complete laptop but costs far more: Macworld's 128GB unit was 6,149 USD as tested, while Tom's Guide puts the 128GB Flow Z13 at over 3,000 USD. [src1, src17, src4]

Pick the ASUS ROG Flow Z13 (2025, 128GB) if: you need Windows or x86 software and want the most GPU-addressable memory for the money.
Pick the Apple MacBook Pro 16 (M5 Max, 128GB) if: you want the faster memory, a full-size screen and keyboard, and long battery life.

Dell Pro Max 18 Plus (RTX PRO 5000) vs Dell Pro Max 16 Plus (RTX PRO 5000)

The same RTX PRO 5000 (24GB). StorageReview measured the 16 Plus at about 96% of the 18 Plus's token rate, at 5.63 lb against 7.17 lb, and 6 hours 21 minutes of battery against 3 hours 39 minutes. [src2] Notebookcheck expects the same CPU and GPU may run faster in the larger chassis, because the GPU in its 16 Plus settled near 125W. [src11]

Pick the Dell Pro Max 18 Plus if: every token per second counts and the laptop will live on a desk.
Pick the Dell Pro Max 16 Plus if: you will carry it.

Decision Logic

If your models are larger than 24GB (70B-class and up)

→ Apple MacBook Pro 16 (M5 Max, 128GB). Its GPU can use up to 128GB at 614GB/s, and Hardware Corner ran a 122-billion-parameter model on the 128GB M5 Max at about 55 to 65 tokens per second. [src7, src6] On Windows, the ASUS ROG Flow Z13 (2025, 128GB) can assign up to 96GB to its GPU. [src17] No discrete GPU on this card has more than 24GB.

If you train or fine-tune with CUDA

→ ASUS ROG Strix Scar 18 (2025, RTX 5090), Tom's Guide's best AI laptop for power users, with a 24GB RTX 5090 at 175W. [src1, src9] If you want a workstation with professional drivers instead, the Dell Pro Max 16 Plus (RTX PRO 5000, 24GB), or the Dell Pro Max 18 Plus, which Dell still sells with that GPU. [src2, src11, src29]

If you want the fastest token rate on models that fit in 24GB

→ Dell Pro Max 18 Plus, StorageReview's fastest local-AI laptop; the Dell Pro Max 16 Plus gives about 96% of it in a lighter laptop with nearly twice the battery life. [src2]

If your organisation requires ECC memory

→ Lenovo ThinkPad P16 Gen 3: ECC system memory up to 128GB, and ECC graphics memory on its RTX PRO GPUs. [src16] The Dell Pro Max 16 Plus has no ECC memory option. [src11]

If you need to work on battery

→ Apple MacBook Pro 16 (M5 Max): 18.5 hours of Wi-Fi browsing in Notebookcheck's test, though full load drains it even on the charger. [src3] The RTX 5090 laptops here lasted from about 1 hour 25 minutes of gaming (Razer Blade 18) to 7 hours 21 minutes of light use (Razer Blade 16). [src19, src20]

If you mostly train in the cloud, or your budget is under $1,500

→ Every laptop on this card costs far more than that; see Best Laptops for Data Science (2026) or Best laptops for software developers (2026).

Default recommendation (unknown requirements)

→ Apple MacBook Pro 16 (M5 Max) with as much memory as you can afford. Its GPU reaches more memory than any other laptop here, it lasts longest on battery, and Notebookcheck rates it 95% for AI performance. [src3, src7] Choose the ASUS ROG Strix Scar 18 (2025, RTX 5090) instead if your tools require CUDA. [src1]

• Two roads to local AI. StorageReview describes discrete RTX GPUs as the fastest per token but limited to what fits in 8GB to 24GB of VRAM, and unified or shared memory designs, led by AMD's Ryzen AI Max+, as trading speed for capacity. Its large-model pick, HP's ZBook Ultra G1a on the Ryzen AI Max+ PRO 395 with 128GB, loaded DeepSeek-R1 70B but generated 8.96 tokens per second on Gemma 3 27B, slower than every RTX PRO laptop it tested on small models. [src2] Apple's M5 Max sits on the capacity side with more bandwidth than AMD's 256-bit LPDDR5X-8000 design: 614GB/s. [src7, src18]
• NVIDIA's RTX Spark laptops are arriving. StorageReview reports that pre-orders opened October 7 for October 16 availability of the first RTX Spark N1X laptops, among them the Dell XPS 16 Creator Edition and the HP OmniBook Ultra 16, with a Blackwell GPU and up to 128GB of unified memory; it has not tested one yet. [src2] Tom's Guide is also waiting for them. [src1]
• More memory for AMD's design. StorageReview expects HP's ZBook Ultra G3a 16, with up to 192GB of unified memory and 160GB of it assignable to the GPU, in October; it has not benchmarked a shipping unit. [src2]
• New MacBook Pros may be weeks away. Bloomberg, as reported by MacRumors, says Apple will introduce redesigned 14-inch and 16-inch MacBook Pro models, thinner and lighter with OLED touchscreens, on or around October 27. [src27] Macworld already suggested waiting if your needs are not pressing. [src4]
• The RTX 5090 premium buys memory more than speed. Ultrabook Review found the RTX 5090 roughly 10% to 15% faster than the RTX 5080 in the Strix Scar 18 for a 25% higher price, and Windows Central, reviewing the Razer Blade 18, says the step above an RTX 5070 Ti pays off only if you use the extra VRAM for local AI. [src10, src19]
• Prices went up in 2026. Tom's Guide reports that Apple raised MacBook prices in June 2026, with the MacBook Pro M5 Pro moving from 2,199 USD to 2,499 USD; Macworld notes that M5 Max models now start with 2TB of storage, at a 200 USD higher entry price. [src1, src4]
• NPUs are not the point yet. StorageReview says most local LLM runtimes lean on the GPU, that NPUs earn their keep on efficiency, and that you should "buy for the GPU and memory first". [src2]

• The Price column shows the Amazon offer our daily check found for each listing we link, where there was one; it is updated with each release. Figures written in USD are as-tested or list prices reported by the outlet named beside them.
• Laptop reviews test one configuration. Where the configuration we link differs from the one reviewed (memory on most of these laptops; the GPU on the MSI Raider 18 HX AI and the Lenovo Legion Pro 7i Gen 10), the section says so.
• A laptop RTX 5090 is not the desktop card: it has 24GB of VRAM against the desktop card's 32GB, and runs at up to 175W against 575W. Hardware Corner's RTX 5090 comparisons use the desktop card. [src9, src6, src19]
• Token rates are not comparable across outlets: StorageReview uses UL Procyon's text-generation test and LM Studio or Ollama, Hardware Corner reports mlx_lm and llama.cpp runs, and Tom's Guide reports Geekbench AI scores. [src2, src6, src1]
• The MacBook Pro 16's cooling holds the M5 Max to about 70W under sustained load, and its 140W adapter cannot keep the battery charged at full load. [src3] The ASUS ROG Flow Z13 runs hot and loud and throttles somewhat. [src17]
• Availability on Amazon changes daily. Statements on this page give the date we checked; the picks table at the top shows today's result.