I’ve been benchmarking computers since my first PC build back in college — a humble Pentium 4 with 512MB of RAM that I was absurdly proud of. Back then, benchmarking was simple: run a single test, get a number, and feel validated that your budget build wasn’t a complete disaster. Fast forward to 2026, and the landscape looks nothing like those early days. Our devices handle AI workloads, 4K video encoding, and real-time ray tracing. The benchmarks we use to measure them need to keep up.
That’s exactly what Primate Labs delivered yesterday with the release of Geekbench 7, the latest version of their cross-platform benchmark suite. It’s available now for Windows, macOS, Linux, Android, and iOS — and it’s still free for personal use. I’ve spent the past 24 hours putting it through its paces across several devices, and I’ve got to say, this update is their most significant in years.
.jpg){kind=link}
What’s New in Geekbench 7
On the surface, Geekbench 7 looks familiar — same clean interface, same single-core and multi-core scores, same cross-platform compatibility. But under the hood, Primate Labs made three substantial changes that genuinely matter for anyone who uses benchmarks to make purchasing decisions.
New Media Workloads
The biggest headline is the addition of dedicated media workloads that model real-world tasks. Geekbench 7 now tests how your CPU handles:
- AV1 screen-sharing encode — modeling what happens when you share your screen in a video call
- Opus audio compression — the same codec used by voice memo and podcast apps
- Whisper speech recognition — decoding audio and generating live captions simultaneously, just like video playback with auto-generated subtitles
These aren’t synthetic abstractions. These are the exact workloads your devices handle daily, especially if you’re working remotely or creating content. If you’ve ever noticed your laptop fan spinning up during a Zoom call, you’ve experienced exactly what these benchmarks now measure.
Alongside the media workloads, Geekbench 7 adds a Game Physics test built on the Jolt Physics engine (used in popular titles like Horizon Forbidden West and Tears of the Kingdom), expands the Photo Editor workload with more realistic edits, and updates the Photo Library to support JPEG XL and DNG imports.
A Smarter Approach to Multi-Core Testing
Here’s the change I appreciate most. In Geekbench 6, every workload ran in multi-threaded mode for the multi-core score. That sounds fair until you realize that not every real-world task benefits from multiple cores. An HTML5 browser test, for example, is essentially single-threaded — browsers just don’t split a page render across 16 cores.
In Geekbench 7, a workload only runs multi-threaded if the real-world task it models actually uses multi-threading. The HTML5 Browser test is excluded from the multi-core suite. The result is a multi-core score that more accurately reflects how your device handles actual applications, not a theoretical maximum that looks impressive in a screenshot but doesn’t translate to real-world performance.
GPU Benchmark Gets a Major Overhaul
Geekbench 7’s GPU benchmark shifts focus to machine learning and content creation — which, let’s be honest, is where GPUs earn their keep in 2026. The new workloads include:
- Face tracking and real-time filters — similar to what social media apps do
- ML-based image upscaling — modeling super-resolution features in tools like Topaz or Adobe
- Background blur for video conferencing — yes, exactly what your webcam does
There are also new GPU image editing and synthesis tests: RAW image processing, LUT-based video color grading, path tracing, and fluid simulation. These are the kinds of tasks that separate a capable GPU from an exceptional one.
And here’s something the community has been asking for: CUDA support. Geekbench 7 now supports CUDA alongside OpenCL, Vulkan, and Metal. If you’re running an NVIDIA GPU, you can finally benchmark using the API that powers its most demanding applications. This alone makes the GPU benchmark significantly more relevant for anyone testing workstation-grade hardware.
How to Download and Install Geekbench 7
Getting started is refreshingly straightforward — a welcome change from some of the complex developer tool setups I’ve covered, like building your first MCP server with FastMCP. The process varies slightly depending on your platform.
Windows
Head to geekbench.com and download the Windows installer. Run the executable — it’s a standard installer that takes about thirty seconds. Once installed, launch Geekbench 7 from the Start menu. The free version gives you full CPU and GPU benchmarking with online result upload. The Pro version (currently 20% off until August 6) adds command-line automation, offline result management, and commercial usage rights.
macOS
Download the DMG from geekbench.com, open it, and drag Geekbench 7 to your Applications folder. First launch may trigger Gatekeeper — right-click and select Open if needed. macOS users get the same feature set as Windows, including Metal GPU benchmarking.
Linux
Primate Labs provides a tarball with a standalone binary. Download it, extract it, and run ./Geekbench7 from the terminal. The Linux version supports both CPU and GPU benchmarking, with Vulkan for GPU testing. If your distro doesn’t have Vulkan drivers installed, grab them via your package manager — on Ubuntu or Debian, that’s sudo apt install mesa-vulkan-drivers.
Android and iOS
Search for “Geekbench 7” on the Google Play Store or Apple App Store. The mobile versions run the same workloads as desktop, which means you can directly compare your phone’s performance against your laptop. I tested it on my daily driver — a mid-range Android phone — and the entire CPU benchmark completed in under five minutes.
Running Your First Benchmark
Once installed, the process is dead simple:
- Launch Geekbench 7 — you’ll see a clean window with two main buttons: Run CPU Benchmark and Run GPU Benchmark
- Close any resource-heavy applications (browser tabs, video players, background compiles)
- Click Run CPU Benchmark — the test takes about 5–8 minutes depending on your hardware
- After completion, you’ll see your single-core and multi-core scores
- Click Run GPU Benchmark — this takes another 3–5 minutes and tests compute, graphics, and ML workloads
- Results automatically upload to the Geekbench Browser (you can opt out if privacy is a concern)
One thing I’d recommend: run the benchmark at least twice and take the average. The first run often warms up the thermal profile, and some devices throttle after extended load. A second run gives you a more consistent baseline — the same principle I covered in my tutorial on secure file transfers with Croc. This is especially true for laptops and phones where thermal constraints can cause score variation of 5–10% between runs.
Understanding Your Scores
Geekbench 7 scores follow the same scale as Geekbench 6, so direct comparisons between versions are possible — though the new workloads will naturally produce different numbers. A score of:
- 1,000–2,000 — Entry-level or older hardware. Fine for basic productivity, not great for media work or gaming
- 2,000–4,000 — Mid-range. Solid everyday performance. Good for light content creation
- 4,000–8,000 — High-end desktop or recent laptop. Handles media encoding, development, and gaming well
- 8,000+ — Workstation or flagship hardware. AI workloads, 4K video editing, and heavy compilation are comfortable
These are rough guidelines — real-world performance depends on your specific use case. A high score won’t fix a poorly optimized application, and a lower score doesn’t mean your machine is useless. What benchmarks give you is a consistent, reproducible way to compare hardware before you buy, or to validate that your system is performing as expected.
I ran Geekbench 7 on my development machine and compared the results against the Geekbench Browser to see where my hardware sits relative to similar configurations. The browser lets you filter by device, operating system, and even specific processor models. It’s a fantastic resource if you’re planning an upgrade and want to see real-world data instead of marketing numbers — the same data-driven approach I explored in my guide on tracking AI API spending with Rekon.
Geekbench 7 Pro: Is It Worth It?
The free version is genuinely useful and likely sufficient for most people. You get full CPU and GPU benchmarking, online result comparison, and cross-platform testing. The Pro version adds:
- Command-line interface — automate benchmarks in CI/CD pipelines or scheduled testing
- Offline results — save and manage benchmark data locally without uploading
- Commercial license — required if you’re benchmarking for work purposes
- Export to CSV/JSON — useful for tracking performance changes over time
At $99 normally (currently $79 with the launch discount), Pro makes sense for IT professionals managing a fleet of machines or developers who want to benchmark in their CI pipeline. For individual enthusiasts? Stick with the free version — it’s already excellent.
Final Thoughts
Geekbench 7 is a meaningful update that reflects how computing has evolved since Geekbench 6 launched. The new media workloads, smarter multi-core methodology, and CUDA support all point in the right direction. Primate Labs didn’t just add more tests — they rethought what a benchmark should measure in an era where AI, video, and real-time collaboration define our daily computing experience.
Whether you’re researching your next laptop upgrade, validating a new PC build, or just curious how your current hardware stacks up, Geekbench 7 gives you a reliable, cross-platform baseline — just one more way to be intentional about the technology you rely on, whether you’re self-hosting your own DNS server or researching your next PC build. Download it, run it, and see where your devices stand. You might be surprised — I certainly was when I discovered my “aging” development machine still holds its own against newer hardware.
The 20% launch discount on Geekbench 7 Pro runs until August 6, 2026, but the free version is all most of us will ever need. Happy benchmarking.