M5 MacBook Air: Is It Actually Worth It for Developers?
M1-M5 MacBook Air benchmarks compared: SSD speed, compile times, thermal throttling, and Wi-Fi 7 tested for real developer workloads.
Drafted with Claude from source material, checked by automated verification, and reviewed before release. How we make these.
What’s actually new in the M5 MacBook Air?
The M5 MacBook Air keeps the same chassis, keyboard, screen, MagSafe connector, and port layout as the M4 before it. What changed is inside: a new N1 wireless chip that adds Wi-Fi 7 support, a jump from Thunderbolt 4 held over from last year (no Thunderbolt 5 here), a storage controller that finally fixes years of slow base-tier SSDs, and a CPU with two “super cores” among its 10 total cores. For developers, the SSD fix and the compile-time gains matter more than the headline chip name.
TL;DR
- SSD speeds on the M5 more than double every prior Air, including the M4, fixing a bottleneck that had persisted since the M1 while base storage jumped from 256GB to 512GB.
- Wi-Fi 7 is present but capped, Apple limits the M5 Air to 160MHz channels instead of the full 320MHz, so real-world throughput lands close to the M4 and even slightly behind the M3 in testing.
- CPU gains are incremental in single-core work, with Speedometer showing just a 5% bump over the M4, the smallest generational jump since the M1 launched.
- Real build workloads show bigger gains, a llama.cpp compile ran about 14% faster than on the M4 and nearly twice as fast as the original M1.
- Thermal throttling is real but modest, the M5 finished a sustained billion-integer merge sort fastest of all five generations but also showed the largest performance drop-off (about 6%) across repeated iterations.
- Memory bandwidth jumped 26% to 142GB/s in STREAM benchmark testing, tracking closely with Apple’s own claimed 28% improvement, which matters for local AI workloads.
- External display support is unchanged from the M4, two external monitors plus the built-in screen, but the M5 can now drive both external displays off a single Thunderbolt port instead of needing both.
How much faster is the M5 than the M4 for coding work?
The answer depends heavily on what kind of “faster” you’re measuring. In Speedometer, a benchmark that simulates responsive web app performance across frameworks (relevant to anyone running Electron-based editors like VS Code all day), the M5 beat the M4 by just 5%. That’s the smallest year-over-year gain since the M1 launched: M1 to M2 delivered a 22% jump, M2 to M3 hit 14%, M3 to M4 came in at 11%. The curve is flattening.
But raw single-core speed isn’t where developers spend most of their CPU cycles. Compiling code, running test suites, and multi-threaded builds lean on all available cores, and that’s where the M5 pulls ahead more convincingly. A single-core C++ sort benchmark (using dynamic core-switching across the M5’s cores) finished about 8% faster than the M4, a real gain since core count didn’t change between the two chips.
The more telling test was an actual llama.cpp build from a fixed commit. Build times across generations: 1 minute 40 seconds on the M1, 1:22 on the M2, 1:14 on the M3, 1:03 on the M4, and under 54 seconds on the M5, roughly 14% faster than the M4 and nearly double the M1’s speed. That’s the kind of gain that shows up in your daily workflow if you’re compiling large codebases repeatedly.
Did Apple fix the MacBook Air’s SSD problem?
Yes, and dramatically. For four straight generations, Apple kept boosting CPU and GPU performance while base-tier SSD speeds stagnated or even regressed. The M2 and M3 base models, shipping with 256GB drives, tested slower on sequential writes than the older M1’s 512GB drive. It was a persistent, well-documented crack in an otherwise improving lineup.
The M5 breaks that pattern hard. Sequential read and write speeds more than double every previous generation tested, including the M4, to the point where an external SSD used for comparison became the slowest drive in the room. Random read/write performance improved too. Apple also dropped the 256GB base configuration entirely, moving the starting point to 512GB, with configurations now scaling up to 4TB (at a price premium). For developers who routinely run out of space installing SDKs, containers, and dependencies, this closes one of the most annoying limitations of the Air lineup.
Is Wi-Fi 7 on the M5 actually faster?
Partially. The M5 introduces Apple’s N1 wireless chip, which supports Wi-Fi 7 features including Multi-Link Operation (MLO) and wider channels, plus a Bluetooth upgrade with device-distance awareness. In practice, testing with iperf3 against a Wi-Fi 7 access point and 10-gigabit wired network told a less impressive story. On the 5GHz band, the M5 topped the chart at just 625 megabits per second, well short of Wi-Fi 7’s potential. On 6GHz, results landed close to the M4 and slightly behind the M3.
Remy doesn't write the code. It manages the agents who do.
Remy runs the project. The specialists do the work. You work with the PM, not the implementers.
The bottleneck traces back to a specific limitation: Apple caps the M5 MacBook Air to 160MHz channels rather than the full 320MHz that Wi-Fi 7 supports. Whether that’s a deliberate segmentation decision to differentiate the Air from higher-tier MacBook Pro models remains unclear, but the practical result is that Wi-Fi 7 branding doesn’t yet translate into a meaningful throughput jump on this machine.
Does the M5 MacBook Air throttle under sustained load?
Somewhat, and it’s worth understanding why. The MacBook Air has no fan, relying entirely on passive cooling, which makes it a useful stress test for how chips behave under sustained multi-core load without active thermal management (unlike the MacBook Pro).
A sustained test running a billion-integer merge sort across five iterations, using every available core, showed the M5 finishing fastest overall at 11 minutes 33 seconds for the full run, comfortably ahead of the M4, M3, M2, and M1. But speed and thermal stability aren’t the same thing. The M5 also showed the largest performance drop across iterations, about 6% slower by the fifth run compared to the first, indicating some throttling under extended load. The M3 and M4 each showed a steadier 5% drop. The M2 was the most thermally consistent (just 3% variance) but also the slowest overall, while the M1, despite being the oldest chip, ran the coolest to the touch throughout testing.
The takeaway: the M5 is faster everywhere, including under sustained load, but it’s not immune to warming up and slowing down slightly during extended, all-core workloads like large compiles or long-running data processing jobs.
Does the extra memory bandwidth matter for local AI?
For anyone running local language models, yes. AI inference workloads depend on two things: raw processing power and memory bandwidth, since token generation is bottlenecked by how fast data moves between memory and the GPU. The STREAM benchmark, a longstanding memory bandwidth test updated to support Apple’s M5, measured 142GB/s on the M5 versus 113GB/s on the M4, a 26% increase that lines up closely with Apple’s own claimed 28% improvement. For developers experimenting with local model inference on Apple Silicon, that bandwidth gain is arguably more relevant day-to-day than the modest CPU single-core improvements.
Frequently Asked Questions
Is the M5 MacBook Air a big upgrade over the M4?
It depends on the workload. Single-core, everyday responsiveness gains are small (around 5% in Speedometer testing), the smallest generational jump since the M1. But multi-core compile tasks, SSD speed, and memory bandwidth all show substantial improvements over the M4.
How much faster is the M5’s SSD compared to the M4?
Sequential read and write speeds on the M5 more than double the M4’s, based on hands-on testing across five generations of MacBook Air. The base storage tier also doubled from 256GB to 512GB.
Does the M5 MacBook Air support full Wi-Fi 7 speeds?
Not fully. While it includes Wi-Fi 7 hardware capable of MLO and 320MHz channels, Apple limits the M5 Air to 160MHz channels, so real-world throughput tested close to the M4 and slightly behind the M3 in some conditions.
Can the M5 MacBook Air handle sustained compile jobs without slowing down?
It handles sustained multi-core loads faster than any previous Air, but it does show some throttling, around 6% performance drop across repeated iterations in stress testing, more than the M4 or M3 showed.
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
How many external monitors can the M5 MacBook Air drive?
The same as the M4: two external displays plus the built-in screen, for three total. The M5’s one improvement is that both external displays can now run off a single Thunderbolt port, freeing up the second port.