Apple’s new M6 chip outpaces last year’s M5 on nearly every spec that matters: a smaller 2nm process, two more CPU cores, a second Neural Engine, and native hardware support for FP8 AI workloads. The gains aren’t marginal either, with Apple citing up to 1.2x faster multithreaded performance and a 30% jump in peak GPU compute for AI tasks. M6 debuts first in the new Mac mini, starting at $899, while M5 remains available in Pro configurations for buyers who need more GPU and memory headroom.
Here’s how the two chips stack up feature by feature:
- Process node: M6 is Apple’s first 2nm chip; M5 was built on a 3nm process
- CPU: M6 has 12 cores (two super cores, four performance cores, six efficiency cores) versus 10 cores on M5
- Neural Engine: M6 has two 16-core Neural Engines running simultaneously; M5 has one
- Memory bandwidth: M6 hits 170GB/s, up from 153GB/s on M5
- AI processing: M6 handles FP8 natively in hardware; M5 processes FP8 through software
- Performance gains: M6 is up to 1.2x faster in multithreaded tasks and offers a 30% increase in peak GPU compute for AI over M5
- First product: M6 debuts in the new Mac mini, available to preorder now and in stores September 22, 2026
- Price: Mac mini with M6 starts at $899 in the US
Apple announced the M6 chip, alongside the M5 Ultra, yesterday. The M6 is arriving as a standalone chip this time, with no M6 Pro or M6 Max variant. Apple is holding those higher-end options back until the M7 family launches in 2027, a departure from the base-plus-Pro-plus-Max pattern it’s followed for every prior M-series generation.
The headline change is architectural rather than physical. Where M5 relied on a mix of performance and efficiency cores, M6 is the first M-series chip to combine all three CPU core types Apple has ever used: super cores tuned for maximum single-threaded performance, performance cores built for demanding multithreaded work, and efficiency cores that handle background tasks. Apple says this heterogeneous layout, absent from M5 entirely, gives M6 the world’s fastest single-threaded performance of any chip it’s shipped.
Chip and Performance: M6 vs M5
M6 outguns M5 on the GPU as well, gaining two more cores to land at 12, and each core now includes a Neural Accelerator alongside an updated shader core architecture, Dynamic Caching, and hardware-accelerated ray tracing. Apple pairs that with a dual 16-core Neural Engine, the first time it’s put two Neural Engines in one chip, delivering up to 2x the peak compute of M5’s single-engine setup, with system frameworks automatically splitting work across both engines at once.
Memory bandwidth rises to 170GB/s from 153GB/s on M5, a 10% gain, and unified memory tops out at 32GB on the base M6. In practical terms, Apple says the M6 Mac mini runs LLM prompt processing in LM Studio up to 13.5x faster than an M1 Mac mini and 4.8x faster than an M4 model. Gaming gets a boost too: Cyberpunk 2077: Ultimate Edition with ray tracing runs up to 2x faster on M6 than on M4.
The M5 Pro configurations sold alongside M6 push further for pro workloads, with up to an 18-core CPU, a 20-core GPU with third-generation ray tracing, and 307GB/s of memory bandwidth across up to 64GB of unified memory, all specs that exceed what the base M6 offers. M6 models use Thunderbolt 4 ports, while the M5 Pro models step up to Thunderbolt 5.
AI
The dual Neural Engine setup isn’t something users toggle manually. Apple’s system frameworks handle the split automatically, routing AI workloads across both 16-core engines at once rather than requiring developers to write separate code paths, a capability M5’s single Neural Engine can’t replicate. Native FP8 support works the same way: apps that lean on lower-precision AI models get the memory and speed benefit in hardware without needing the software workaround M5 required.
| Feature | M5 | M6 |
|---|---|---|
| Process node | 3nm | 2nm |
| CPU cores | 10 (mix of performance and efficiency) | 12 (2 super, 4 performance, 6 efficiency) |
| GPU cores | 10 | 12 |
| Neural Engine | Single 16-core engine | Dual 16-core engines |
| Memory bandwidth | 153GB/s | 170GB/s |
| Max unified memory | Unknown (not specified in brief for base M5) | 32GB |
| FP8 processing | Software-based | Native hardware |
| Multithreaded performance | Baseline | Up to 1.2x faster than M5 |
| Peak GPU AI compute | Baseline | 30% increase over M5 |
| First shipping product | MacBook Pro, MacBook Air, iPad Pro, Vision Pro (Oct 2025) | Mac mini (Sept 22, 2026) |
Frequently asked questions
What is the main difference between M6 and M5?
M6 uses a 2nm process with a 12-core CPU, a 12-core GPU, and two 16-core Neural Engines, while M5 uses a 3nm process with a 10-core CPU, a 10-core GPU, and a single Neural Engine.
How much faster is M6 than M5?
Apple says M6 delivers up to 1.2x faster multithreaded performance and a 30% increase in peak GPU compute for AI compared to M5.
When does the M6 chip launch?
The M6 chip debuts in the new Mac mini, available for preorder now and launching in stores on September 22, 2026.
Will there be an M6 Pro or M6 Max?
No. Apple is releasing M6 only as a standalone chip and will hold off on Pro and Max variants until the M7 series arrives in 2027.
What does M6’s native FP8 support mean?
M6 processes FP8, a data format used in AI workloads, directly in hardware, while M5 required software to handle the same task, making M6 faster and more memory-efficient for AI.
Should you wait for M6?
If you’re buying a Mac mini today and don’t need Pro-level GPU power, the $899 M6 model is the better long-term buy given its 2nm efficiency and dual Neural Engine. Anyone currently using an M5 machine bought in the past year has little reason to switch, but M4 or older owners eyeing a MacBook Pro should note the base M6 version arrives in October 2026 with no Pro or Max option until 2027, so those needing more GPU headroom will have to look at current M5 Pro or M5 Max configurations instead.
