New Mac mini and Mac Studio arrive with AI-focused upgrades
Apple has launched a new generation of Macs, including the first Mac mini with the M6 chip and an upgraded M5 Pro model, alongside a revamped Mac Studio lineup powered by the M5 Max and M5 Ultra. Pre-orders for the Mac mini models are now going live at Amazon, as reported by 9to5Toys.
The M6 Mac mini starts at $899 for the 16GB/256GB configuration, with the 16GB/512GB version at $1,099 and the 24GB/512GB at $1,299. The M5 Pro Mac mini, available in a 24GB/512GB configuration, is priced at $1,699. 9to5Toys notes that the M6 Mac mini uses a 2nm process, offers 170GB/s of memory bandwidth, and caps at 32GB; the M5 Pro version has 307GB/s and caps at 64GB. Both models are currently available for pre-order on Amazon.
For those looking to upgrade, 9to5Toys reports that Amazon is not offering trade-ins for the new models, but the official Apple site provides trade-in credits: a 24GB/512GB M4 Mac mini can be worth up to $525 off the new M6 Mac mini, potentially bringing the price down to as low as $374 with a mid-range M4 trade-in. The retailer also highlights the availability of the Apple Polishing Cloth for $9, which it suggests buyers should snag before it sells out.
9to5Toys also notes that stock of the previous-generation M4 Mac mini had largely disappeared from retail over the past few months, and it's unclear whether Apple will be able to keep the new models in stock, citing past supply issues with the 'Neo' product.
Mac Studio: a local alternative to AI subscriptions?
Beyond the Mac mini, Apple has announced new Mac Studio models powered by the M5 Max and M5 Ultra chips, which the company positions as a way to run large AI models on-device without counting tokens or worrying about cloud costs. Forbes dives into whether this claim holds up financially, asking if the new Mac Ultra can replace $200-per-month AI coding subscriptions.
Forbes reports that the Mac Studio with the M5 Max starts at $2,499 with 36GB of unified memory and up to 614GB/s of bandwidth, while the M5 Ultra starts at $5,499 with 96GB and 1.2TB/s, a 50% bandwidth jump over the M3 Ultra. The top-end M5 Ultra configuration with 256GB of memory and 16TB of storage costs $18,299, with a 512GB configuration arriving in late October at an unannounced price. Both initial models ship on September 22.
According to Forbes' analysis, a 128GB M5 Max Mac Studio costs about $4,800, and a 256GB M5 Ultra costs $9,499. Assuming eight hours of electricity per day at the U.S. average rate of 18.44 cents per kilowatt-hour, running the Mac Studio costs roughly $10 per month in electricity, or about $36 per month if run 24/7.
Forbes compares these hardware costs against popular AI subscriptions: Claude Max is priced at $100 or $200 per month, ChatGPT Pro is priced identically, and GitHub Copilot Max is $100 per month. The analysis suggests a 128GB M5 Max Mac Studio would break even against a $200/month subscription in about two years, while a 256GB M5 Ultra would take about four years. The base $2,499 Mac Studio, meanwhile, would break even in 12.5 months.
Anthropic's documentation, cited by Forbes, states that Claude Code costs "around $13 per developer per active day and $150-250 per developer per month" for metered API spend in enterprise deployments.
Performance and model compatibility
Forbes also examines whether the new Macs can actually run the latest open-weight models. It notes that Moonshot's Kimi K3, with 2.8 trillion parameters, does not fit on the 512GB Mac Studio, while DeepSeek's V4-Pro-0813 checkpoint is 893GB. Alibaba's flagship Qwen3.8 has 2.4 trillion parameters, but the Qwen3-Coder-Next, an 80B mixture-of-experts model, fits in 128GB, as does OpenAI's gpt-oss-120b.
Benchmark data from Terminal-Bench 2.1, as reported by Forbes, shows Claude Code with Fable 5 scoring 83.8%, while GLM-5.1 scores 58.7% and ranks 17th. GLM-5.1 is a 754B model that needs roughly 420GB to hold at 4-bit. On Terminal-Bench 2.0, Qwen3-Coder-Next driven by Claude Code scores 30.9%, against Claude Opus 4.5's 53.9%.
Forbes also relays developer reports that a 16,000-token system prompt took about 90 seconds on a 512GB M3 Ultra, while a 48GB M4 Pro laptop measured time-to-first-token at 20 seconds cold and about a minute deep into a coding session. Apple's own benchmarks, cited by Forbes, show the M5 MacBook Pro improves time-to-first-token 3.3x to 4.1x compared to the M4 and improves token generation 19-27%. Apple claims up to 4x faster LLM prompt processing on the M5 Ultra versus the M3 Ultra.
Perspectives
Local AI advocate
Bart de Witte, CEO of Berlin-based Isaree, told Forbes that the M5 Ultra "holds a frontier-class medical AI model, the kind that today 'requires' a data center, entirely in local memory" and that "no patient data leaving the building," highlighting the privacy benefits of on-device AI for sensitive fields like healthcare.
Cost-conscious developer
Forbes' analysis, while acknowledging the appeal of owning local hardware, questions the financial case for most users, suggesting that for developers who don't hit rate limits or face massive API bills, the upfront cost of a Mac Studio may not justify replacing a $200/month subscription.