An engineering sample believed to be AMD's next-generation mobile CPU "Medusa Point" has recorded a single-core score of 3329 and a multi-core score of 16555 on Geekbench 6. Compared with the average registered results for the Ryzen AI 9 365, a 10-core Zen 5-generation product, the differences reach 34.8% and 33.4% respectively. These figures raise expectations for the next-generation CPU. However, AMD has not disclosed the identity of the sample or confirmed the adoption of Zen 6. The prototype CPU with the same model number and configuration had already gained as much as 9.7% in score compared to a measurement taken 9 days earlier, meaning there is insufficient test data to definitively establish a generational difference.
A 10-Core Configuration Appears on Plum-MDS1
A result registered on Geekbench Browser on July 17, 2026 identifies the CPU as "AMD Eng Sample: 100-000001713-33_N," with the system and motherboard identified as "AMD Plum-MDS1." The OS was Windows 11 Pro, the power plan was set to "Balanced," and the measurement was taken using the Windows AVX2 version of Geekbench 6.6.0. The result was judged valid, recording a single-core score of 3329 and a multi-core score of 16555.
According to the registration data, the 10 cores/20 threads are divided into two clusters of 4 and 6 cores, with 1MB of L2 cache per core and 32MB of L3 cache. The instruction set also includes AVX-512 FP16. AMD's commercially available Ryzen AI 9 365 is also a 10-core/20-thread product combining 4 Zen 5 cores with 6 Zen 5c cores, with a total of 10MB of L2 cache. The core configurations are very similar.
However, the "valid" designation in Geekbench is a judgment about the benchmark result itself, and does not mean that AMD has certified the product name or architecture. While the board name "Plum-MDS1" and the prototype CPU's model number lead to speculation that this is a Medusa Point candidate, there is no notation of "Medusa Point" or "Zen 6" anywhere in the database. The 32MB L3 cache also cannot be treated as a confirmed specification for an unannounced chip.
A 33.4% Gap Revealed Through Same-Core Comparison
The average of Geekbench 6 registered results for the Ryzen AI 9 HX 370 is 2606 single-core and 13400 multi-core. This prototype CPU exceeded these figures by 27.7% and 23.5% respectively. While the HX 370 is a top-tier product in the current Strix Point lineup, it is a 12-core/24-thread chip combining 4 Zen 5 cores with 8 Zen 5c cores—2 cores more than the 10-core prototype. Although the comparison with this flagship product is eye-catching, the mismatched core counts make it unsuitable as a benchmark for examining generational design differences.
Using the same 10-core/20-thread Ryzen AI 9 365 as the baseline, the comparison looks as follows.
| Geekbench 6 | Single-Core | Multi-Core | Cores/Threads |
|---|---|---|---|
| AMD Plum-MDS1 Prototype CPU | 3329 | 16555 | 10/20 |
| Ryzen AI 9 365 Registered Average | 2469 | 12411 | 10/20 |
| Prototype CPU Difference | +34.8% | +33.4% | Same |
The fact that multi-core performance is 33.4% higher with the same core count provides more meaningful material for considering generational design improvement than the comparison with the HX 370. The overall Geekbench 6 score is built from multiple workloads including compression, web rendering, and code compilation, with machine learning and image processing loads also factored in.
Even so, what this table shows is a difference in scores collected under different environments. The aggregate figures for the Ryzen AI 9 365 are drawn from multiple results submitted by users, and according to AMD's specifications, the default TDP is 28W with a configurable range as wide as 15–54W. Since the power consumption of the Plum-MDS1 was not recorded, even with the same core count, the power conditions cannot be assumed to be identical. This cannot simply be translated into a rate of IPC improvement or performance efficiency.
A Score That Moved by Up to 9.7% Over 9 Days
The same combination of "100-000001713-33_N" and Plum-MDS1 had also been measured on Geekbench 6.6.0 on July 8. That earlier result showed a single-core score of 3174 and multi-core score of 15092, meaning the July 17 measurement improved by 4.9% and 9.7% respectively. Everything from the CPU model number to the core configuration, cache capacity, and installed memory capacity was identical.
The reason behind this 9-day difference cannot be determined from publicly available information. Test conditions ranging from firmware to cooling and memory settings have not been disclosed, and the power limit is also unknown. It cannot be determined whether the prototype underwent further tuning or whether the measurement conditions simply differed. This is precisely why a single top score cannot be regarded as representative of a mass-produced product.
Inconsistencies also remain in the displayed operating clocks. The latest result page records a base frequency of 2.00GHz and a measured maximum frequency of 2152MHz, which does not support any claim that the chip operated in the 5GHz range. Since the actual boost behavior cannot be reconstructed from the measurement page, it is also impossible to separate the score improvement into contributions from higher clock speeds versus architectural improvements.
AMD Has Confirmed Medusa, But Not Zen 6
At its Financial Analyst Day in November 2025, AMD disclosed that its next-generation client CPU roadmap includes "Gorgon" and "Medusa." However, the CPU architecture and core count for Medusa were not disclosed. Clock speeds, TDP, release timing, and product names also remain unannounced. The association of this Plum-MDS1 sample with Medusa Point, and its further identification as a Zen 6 product, remains speculation based on leaked information.
Even so, this record carries two implications: a 10-core AMD prototype CPU exceeded the average of current same-core products by 34.8% in single-core and 33.4% in multi-core performance, and a prototype CPU with the same model number and configuration has repeatedly appeared on Geekbench with updated scores. These are strong signs that development of the next-generation mobile CPU is progressing. However, they are not evidence that definitively establishes the product's generation and performance.
The assessment will change once AMD officially announces the product and multiple measurements emerge from mass-produced laptops under matched power consumption and memory conditions. If a comparable gap is maintained at that point, it can be concluded that the Medusa generation achieved a significant boost in CPU performance without increasing the core count.