After weeks of rumors, several corporate statements have finally put the pieces together: Intel prepares Arrow Lake Refresh for its desktop platform and will do so as a revision focused on polishing performance and stability compared to current models. What was aimed at the end of 2025 has officially moved to a broader window, with arrival in the first half of 2026.
According to the company, this review is not born to enter into a direct war with the competition on all fronts, but to strengthen the desktop catalog and fill specific gaps. In the words of its spokespersons, it will serve to "kick-start the process" on the desktop before the big leap they plan for the next generation.
Launch and platform: when it arrives and where it fits
The calendar places Arrow Lake Refresh in the first six months of 2026, likely to be introduced soon this year to separate it from the next batch of CPUs. It wouldn't be unusual to see it appear at a major industry event, like CES, although that detail has yet to be confirmed.
On the technical side, Intel will maintain the current ecosystem: socket LGA1851 and familiar chipsets (Z890, B860, H810), making life easier for those already invested in this platform. This fits with the idea of ​​an evolutionary revision, rather than a disruptive leap.
Regarding the commercial name, everything indicates that it will follow the line of the current models: Core Ultra Series 2 (Core Ultra 200) for Arrow Lake Refresh, while Panther Lake will remain as Core Ultra Series 3. In this way, the nomenclature would not overlap with future families.
The roadmap also talks about model cadence: it is possible that the K/KF variants have priority release for enthusiasts, and that other more conventional references appear later, even in 2027, as the cycle is completed.
What's changing inside: planned improvements and first clues
Sources agree that we are facing a low-risk refresh: the same CPU and GPU architectures as Arrow Lake-S and identical core counts, with firmware and frequency tweaks to squeeze performance where there's room. This approach fits with the goal of fixing what didn't quite work in the first batch.
Where a clear leap is expected is in the AI ​​part. The new NPU, referred to in leaks as NPU4 with around 48 TOPS, would exceed the 40 TOPS threshold required by features like Microsoft Copilot+, enabling more scenarios local acceleration In day to day.
Among the early evidence, a supposed has appeared Intel Core Ultra 7 365K (engineering sample) on Geekbench 6: 8 P-cores (Lion Cove) and 12 E-cores (Skymont) for a total of 20, without Hyper-Threading and with reported frequencies of up to 5,4 GHz on P-cores. 36 MB of L2 and 30 MB of L3 are quoted, and initial scores of 2.140/19.744 (mono/multi). As this is a sample, does not reflect final performance, but it serves to outline that we are talking about a moderate evolution.
In compatibility, nothing changes: plates with Z890, B860 and H810 should be the natural destiny of these chips, maintaining the LGA1851 + Arrow Lake Refresh combination without any problems for the user who updates.
Looking ahead, Arrow Lake Refresh will act as bridge to Nova Lake. Intel plans to introduce major architectural and process changes (18A), new configurations and a new socket (LGA1954), with a planned release at the end of 2026 and a deployment that would extend through 2027. Hence, this review's mission is to hold the desk until a replacement arrives.
Intel's desktop plan involves a intermediate setup that preserves the platform and names, raises the bar for the NPU, and polishes performance where possible, leaving major changes for the next generation. For those with an LGA1851 motherboard looking to stretch their system, it's a logical short-term upgrade option.