The comparison changes with the job.
The Prestige 16 AI+ gives me a large 120Hz OLED touchscreen, Windows compatibility and 32GB of memory. Against my M5 Air, the result changes with the workload and with how the laptop is powered.
MSI Prestige 16 AI+
Core Ultra X7 358H · Arc B390
32GB LPDDR5X · 1TB SSD
16-inch 2880×1800 OLED touch · 120Hz
M5 MacBook Air
32GB unified memory · 10-core GPU
Fanless macOS laptop
One tested unit of each machine
These results come from the August 26 comparison. Operating systems, graphics backends and some benchmark dates differ. Each selector carries the conditions for the result it shows.
Plugged in, the MSI can sustain more CPU work.
The MSI wins the published ten-minute Cinebench run while plugged in. The Air leads the shorter Geekbench tests. That is a reason to separate workloads, rather than average unrelated benchmark points into one winner.
Sustained CPU · ten-minute run
Cinebench 2026.1.3 multi-core score · Higher is betterPublished card at 2:19: plugged in, sustained run; MSI High Performance GUI on August 4 versus M5 CLI on August 3. One unit each. The card calls the result start-state sensitive and notes an earlier MSI score of 3,813. Do not swap that older score into this pair.
All measurements and test conditions
| Test | Results | Unit and conditions |
|---|---|---|
| Sustained CPU · ten-minute run | MSI Prestige 16 AI+ · 32GB: 3,487 M5 MacBook Air · 32GB: 3,235.94 | Cinebench 2026.1.3 multi-core score Published card at 2:19: plugged in, sustained run; MSI High Performance GUI on August 4 versus M5 CLI on August 3. One unit each. The card calls the result start-state sensitive and notes an earlier MSI score of 3,813. Do not swap that older score into this pair. |
| Short CPU test · single-core | MSI Prestige 16 AI+ · 32GB: 2,689 M5 MacBook Air · 32GB: 4,117 | Geekbench 6.7.1 single-core score Published card at 2:07: plugged in, August 3, one run. Windows and macOS binaries differ. This single-core result is a separate workload from sustained Cinebench. |
| Short CPU test · multi-core | MSI Prestige 16 AI+ · 32GB: 15,643 M5 MacBook Air · 32GB: 16,825 | Geekbench 6.7.1 multi-core score Same 2:07 card and plugged-in conditions. Different test length and workload explain why the ordering need not match ten-minute Cinebench; these scores are not FPS. |
The battery result has a power bill.
On battery, the MSI’s Steel Nomad Light score was 1,212 versus 1,051. Its reported draw was also much higher: 31.7 W versus 15.3 W. A score advantage and a power advantage are different things.
On battery · graphics score
3DMark Steel Nomad Light score · Higher is betterPublished card at 3:03: August 4, one run each, on battery. MSI result was hand-recorded with no result file; the M5 has a saved result. Steel Nomad Light, not regular Steel Nomad. Points are not a game’s FPS.
All measurements and test conditions
| Test | Results | Unit and conditions |
|---|---|---|
| On battery · graphics score | MSI Prestige 16 AI+: 1,212 M5 MacBook Air: 1,051 | 3DMark Steel Nomad Light score Published card at 3:03: August 4, one run each, on battery. MSI result was hand-recorded with no result file; the M5 has a saved result. Steel Nomad Light, not regular Steel Nomad. Points are not a game’s FPS. |
| On battery · reported system power | MSI Prestige 16 AI+: 31.7 W M5 MacBook Air: 15.3 W | Reported watts during the graphics test Power labels from the same 3:03 card. Different operating-system sensor paths make this a directional system comparison, not a calibrated wall-meter or chip-efficiency test. Instantaneous power is not battery runtime. |
The review describes a separate reversal in sustained CPU performance when unplugged. The chart here stays with the battery graphics result actually shown in this passage; it does not convert points per watt into a runtime estimate.
Editing and local AI point in opposite directions.
The Air is well ahead in the Resolve Basic test. The MSI is well ahead generating the selected Qwen model. Pick the job you actually do and keep the software path attached to the result.
Video editing · Resolve Basic
PugetBench overall Basic score · Higher is betterPublished card at 4:17: PugetBench for DaVinci Resolve 2.1.0, Basic preset, Resolve Studio 21, plugged in, one valid run per machine. Basic omits the GPU-effects block; this is not an export-time measurement or every editing workload.
All measurements and test conditions
| Test | Results | Unit and conditions |
|---|---|---|
| Video editing · Resolve Basic | MSI Prestige 16 AI+ · 32GB: 44,818 M5 MacBook Air · 32GB: 74,204 | PugetBench overall Basic score Published card at 4:17: PugetBench for DaVinci Resolve 2.1.0, Basic preset, Resolve Studio 21, plugged in, one valid run per machine. Basic omits the GPU-effects block; this is not an export-time measurement or every editing workload. |
| Local AI · generating the answer | MSI Prestige 16 AI+ · 32GB: 38.25 M5 MacBook Air · 32GB: 20.86 | Generation tokens per second Published card at 4:38: Qwen3.6-35B-A3B UD-Q2_K_XL, TG128, full-offload intent. MSI Vulkan build 10093 versus M5 Metal build 10090. Model and workload match; backend and build do not. This is not the Llama NPU workload in the X2 review. |
A bigger adapter did not settle every power problem.
The charger is part of this review. In one 29.6-minute gaming session, the source reports a 60 W system-power ceiling and 0.562 Wh of battery energy lost while plugged in.
A 140 W adapter removed that wall-power ceiling in later sessions. It did not eliminate a separate intermittent processor-power drop. Those sessions were not matched frame-rate runs, so the larger adapter is not a proven FPS fix.
Choose the workload, then check the power conditions.
The MSI makes sense in my test cycle for its display, Windows software and the local-model result. The Air remains the quieter, lighter option and was stronger in this Resolve run. Neither conclusion requires pretending that the charger behavior was consistent across every session.
The video’s $1,399 Costco purchase price followed a rebate that has expired. Treat it as what I paid, not current shopping advice.
What would change this verdict?
Repeated tests at controlled charge levels, a matched adapter comparison and a firmware update could change the power story. The CPU, graphics and AI results each need their own repeat runs; changing one score does not settle the whole comparison.
Watch the complete review
The short clips above support individual findings. The full video includes narration and the rest of the testing.
Open the full review on YouTube ↗ · Subscribe for the next review ↗
Published-source snapshots: msi. Chart values and conditions.
The rest of this laptop cycle
New benchmark results by email
Get raw benchmark tables and thermal/power test notes when new machines are tested.
Occasional emails when there is something new. No spam, unsubscribe anytime · Privacy
You’re in. New results will come from newsletter@blendlogic.com.
That did not go through. Check the address and try again.