Apple launches M6 iMac and 14-inch MacBook Pro with GPU Neural Accelerators for faster on-device AI
Apple on October 27, 2026 introduced the first two Macs built on its M6 chip, a new 24-inch iMac and a new 14-inch MacBook Pro, and it chose to sell both largely on local AI performance. In each machine the M6 combines a CPU and a GPU of up to 12 cores apiece, two more of each than the chips they replace, a Neural Accelerator inside every GPU core, a dual 16-core Neural Engine and unified memory bandwidth of up to 170GB/s. Apple's headline AI figure for the MacBook Pro is up to 13x faster time to first token when running a large language model compared with the 13-inch MacBook Pro with M1, while for the iMac it claims up to 12x faster prompt processing in LM Studio than the M1 generation and up to 5x faster on-device AI than the M4 iMac it replaces.
Both computers can be pre-ordered from October 28 on apple.com and in the Apple Store app in 35 countries and regions, including the United States, and they reach customers and stores on Friday, November 6. The iMac starts at $1,499 and the 14-inch MacBook Pro at $1,999 in the U.S. Both ship with macOS 27, which carries the current generation of Apple Intelligence and the Siri AI beta that Apple extended to more languages in the 27.2 release.
The launch matters beyond the usual annual refresh because Apple is now quoting language-model metrics, rather than only generic "machine learning" multipliers, as the main measure of a consumer Mac. The figures are Apple's own, measured in its labs in September 2026, and the company does not name the models, sizes or quantization levels behind them, so they describe the direction and rough scale of the improvement rather than a benchmark anyone can reproduce today.
··········
WHAT THE M6 CHIP ADDS FOR AI WORKLOADS.
The architectural change that drives the AI claims is the Neural Accelerator placed inside each GPU core. Apple first moved matrix-multiplication hardware into the GPU with the previous generation of Apple silicon, and the M6 keeps that design while raising the core count to 12 in its top configuration. Matrix multiplication dominates the prefill phase of a language model, the stage in which the model reads the whole prompt before producing its first output token, which explains why Apple's chosen metrics are prompt processing and time to first token rather than raw generation speed.
The Neural Engine is a separate block. Apple describes a dual 16-core Neural Engine that powers Apple Intelligence features with an emphasis on energy efficiency, and it says this unit is up to 1.9x faster than the one in the M5 MacBook Pro for tasks such as image generation, Super Resolution and AI upscaling. For the iMac, whose previous generation used the M4, Apple quotes up to 2x faster AI compute from the Neural Engine. In practice, the system features built into macOS lean on the Neural Engine, while third-party local-model runtimes such as LM Studio rely mostly on the GPU and its accelerators.
Memory is the third variable. The M6 offers unified memory bandwidth of up to 170GB/s, compared with 153GB/s on the M5 MacBook Pro, an increase of about 11%. The 14-inch MacBook Pro starts with 16GB of unified memory and can be configured with up to 32GB. Apple's iMac announcement does not list the memory options, beyond confirming that the M4 iMac used as its comparison baseline had 32GB.
........
Specification | iMac with M6 | 14-inch MacBook Pro with M6 |
CPU | Up to 12 cores | 12 cores |
GPU | Up to 12 cores, Neural Accelerator in each | 12 cores, Neural Accelerator in each |
Neural Engine | Dual 16-core | Dual 16-core |
Memory bandwidth | Up to 170GB/s | Up to 170GB/s |
Unified memory | Options not listed by Apple | 16GB, up to 32GB |
Display | 24-inch 4.5K Retina, 500 nits | 14-inch Liquid Retina XDR, up to 1600 nits HDR |
Wireless | Wi-Fi 7, Bluetooth 6 | Wi-Fi 7, Bluetooth 6 |
Starting price (U.S.) | $1,499 | $1,999 |
........
··········
APPLE'S AI PERFORMANCE CLAIMS AND THEIR BASELINES.
Apple publishes a long list of multipliers, and they only make sense when read against the machine each one is compared with. The MacBook Pro figures use two baselines: the 13-inch MacBook Pro with M1, a five-year-old machine that many buyers still own, and the 14-inch MacBook Pro with M5 configured with a 10-core CPU, 10-core GPU, 32GB of memory and a 4TB SSD. The iMac figures compare with the M1 generation and with the M4 iMac configured with a 10-core CPU, 10-core GPU, 32GB of memory and a 2TB SSD.
........
Machine | Apple claim | Baseline |
14-inch MacBook Pro M6 | Up to 13x faster LLM time to first token | 13-inch MacBook Pro M1 |
14-inch MacBook Pro M6 | Up to 1.3x faster LLM time to first token | 14-inch MacBook Pro M5 |
14-inch MacBook Pro M6 | Up to 25% faster on-device AI workloads | 14-inch MacBook Pro M5 |
14-inch MacBook Pro M6 | Up to 1.9x faster Neural Engine | 14-inch MacBook Pro M5 |
iMac M6 | Up to 12x faster prompt processing in LM Studio | M1 generation |
iMac M6 | Up to 5x faster on-device AI | iMac M4 |
iMac M6 | Up to 2x faster Neural Engine AI compute | iMac M4 |
........
The generation-over-generation gain for the laptop is therefore modest: the time-to-first-token improvement over the M5 model is up to 1.3x, consistent with the 25% figure Apple gives for on-device AI workloads in general. The double-digit multipliers come from comparisons with M1, which predates the GPU Neural Accelerators entirely. The iMac's larger 5x figure has a simple explanation, since the desktop skipped the M5 generation and its baseline is the M4 iMac, which also lacked accelerators in the GPU.
Apple also says the M6 has "the world's fastest CPU core," based on its October 2026 testing against shipping competitive systems and selected industry-standard benchmarks, without publishing the scores. On graphics, the MacBook Pro is rated up to 30% faster than the M5 model, and Apple quotes up to 11.7x faster Blender ray-traced rendering and up to 2.5x faster Xcode builds than the M1 model.
··········
WHAT THE NUMBERS MEAN FOR RUNNING LOCAL LLMS.
Time to first token is the delay between submitting a prompt and seeing the first word of the answer. It grows with prompt length, so it is the number that hurts most when a local model is asked to read a long document, a large code file or a long chat history. A faster prefill phase makes retrieval-augmented setups, local coding assistants and document summarization feel considerably more responsive, which is exactly where Apple's claims are concentrated.
Generation speed, the rate at which tokens appear once the answer has started, depends much more on memory bandwidth, because the model's weights must be read from memory for every token produced. Here the M6's improvement over the M5 is about 11%, from 153GB/s to 170GB/s. Apple does not publish a tokens-per-second figure for generation, so buyers should not expect the 1.3x prefill gain to translate one-for-one into faster streaming output. That reading is an inference from the published specifications, not an Apple claim.
The 32GB ceiling also shapes what can run. Unified memory is shared between macOS, open applications and the model, and the weights alone of a model take roughly its parameter count multiplied by the bytes stored per parameter. As an illustration, a 30-billion-parameter model quantized to 4 bits needs about 15GB for weights before any context cache, which fits comfortably in 32GB; a 70-billion-parameter model at the same precision needs about 35GB and does not. Users who want to run the largest open-weight models locally will still need the higher-memory Pro and Max class machines rather than these entry M6 systems.
Apple does not name any specific model in either announcement. The iMac release cites LM Studio and lists AI-oriented apps including Topaz Studio, Pixelmator Pro and Claude, while the MacBook Pro release says the chip can run LLMs locally while keeping data on the device.
··········
APPLE INTELLIGENCE AND SIRI AI ON THE NEW MACS.
Both machines run macOS 27, and Apple uses them to showcase its current system AI features. Siri AI can draw on personal context across messages, emails and photos, answer questions and take actions across apps, and it is built into Spotlight and context menus. Visual Intelligence is available through a keyboard shortcut, and writing tools and Siri work almost anywhere the user types; in Mail and Messages, Siri can match the user's style and tone. Safari gains tools that group tabs by topic, watch pages for changes such as restocks or price drops, and create custom extensions from a plain-language description, while Shortcuts can build automations from a description and Photos adds intelligent editing.
The fine print limits who gets all of this. According to Apple's footnotes, Siri AI is in beta in macOS 27.2 and supports six languages, and it is not available to users under 13. The same footnote states that Siri AI is unavailable in the EU on iOS, iPadOS and watchOS; macOS is not on that list, a distinction that matters for European buyers and follows the regional restrictions covered in our report on Apple's Siri AI expansion in iOS 27.2. Server-based features, including Siri AI and intelligent photo editing, carry daily usage limits, and Apple says expanded access will be offered for a fee in the future without giving a price or date.
The AI push extends to Apple's own software. The company recently announced AI assistants for its creative and productivity apps, described in our coverage of Apple Creator Studio for Logic Pro, Keynote, Pages and Numbers, and those workloads are the kind the new Neural Engine and GPU accelerators are designed to serve.
··········
AVAILABILITY, CONFIGURATIONS AND PRICING.
The iMac comes in two chip configurations. The model with a 10-core CPU and 10-core GPU costs $1,499, or $1,449 with education pricing, and the model with a 12-core CPU and 12-core GPU costs $1,699, or $1,599 for education. The 12-core iMac adds 2.5 Gigabit Ethernet, Thunderbolt 4 support on all four of its ports, Touch ID as standard and a nano-texture display option. The iMac is offered in blue, green, pink, yellow, purple, orange and silver.
The 14-inch MacBook Pro starts at $1,999, or $1,899 with education pricing, in space black and silver. Apple's release gives only the starting price; other configurations, including the upgrade to 32GB of unified memory, are priced on Apple's online store. The laptop is rated for up to 24 hours of battery life, which Apple calls the longest of any Mac, and it can charge to 50% in about 30 minutes with a compatible higher-wattage USB-C adapter such as the 96W model. Ports include HDMI, MagSafe 3, three Thunderbolt 4 ports, an SDXC slot and a headphone jack.
........
Model | Configuration | U.S. price | Education price |
iMac with M6 | 10-core CPU, 10-core GPU | $1,499 | $1,449 |
iMac with M6 | 12-core CPU, 12-core GPU | $1,699 | $1,599 |
14-inch MacBook Pro with M6 | 12-core CPU, 12-core GPU, 16GB | From $1,999 | From $1,899 |
........
For buyers deciding whether to upgrade for AI work, the arithmetic depends on the machine they own. Owners of M1 Macs gain the GPU Neural Accelerators and, by Apple's figures, an order-of-magnitude improvement in prompt processing. Owners of an M5 MacBook Pro gain up to 1.3x faster time to first token, about 11% more memory bandwidth and the same 32GB maximum memory, which is a smaller step. M4 iMac owners sit in between, with Apple's up-to-5x on-device AI figure reflecting the arrival of accelerators in the GPU.
··········
WHAT IS ESTABLISHED AND WHAT REMAINS UNCERTAIN.
Apple's official announcements establish the core facts: the M6 brings 12-core CPU and GPU configurations with a Neural Accelerator in every GPU core, a dual 16-core Neural Engine and up to 170GB/s of memory bandwidth to the iMac and the 14-inch MacBook Pro; both machines go on pre-order on October 28 and ship on November 6; and the U.S. starting prices are $1,499 and $1,999. The baselines behind every performance claim are documented in Apple's footnotes, and the gains over the immediately preceding chips are measured in tens of percent rather than multiples.
What remains open is the independent picture. Apple has not disclosed which language models, parameter counts or quantization levels it used for its time-to-first-token and LM Studio results, nor has it published token-generation speeds, so the real-world experience with specific open-weight models will only become clear once third-party testing begins after the machines ship. The pricing and timing of paid expanded access to server-based Apple Intelligence features is also unannounced, as is any change to Siri AI availability for EU users on the platforms where it is currently restricted.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
