Improved the feedback form: tap outside a text field to dismiss the keyboard. Improved update reminders of the app to check your App Store region, appear after launch, and remind you again after 24 hours if you choose “Not Now.”

Share your thoughts with the new Send Feedback feature in Settings. Suggest an improvement or report a problem, and optionally leave your email address if you’d like a reply.

Improved usability and clearer benchmark results. • Refreshed dashboard with expandable latest-run details. • Added nearby device comparisons for CPU and AI benchmarks. • Improved AI model selection, download controls and run history. • Clearer upload status and links to online results. • Updated iPhone and iPad model recognition. • Fixed AI result restoration and memory-test cancellation. • Improved readability and accessibility.

New: Memory bandwidth The CPU tab now has 3 pages: Compute, RAM CPU and RAM GPU. The last two measure how fast your device streams data out of memory, once from the processor cores and once from the graphics chip. This is the main limit on on-device AI speed. Generating each token re-reads the whole model, so a model twice the size runs roughly half as fast on the same phone. - Bug fix: stopping a benchmark no longer discards a run that had already finished

• The model browser now lists only models that can actually be benchmarked: embeddings, speech and other non text-generation models are filtered out, and the search explains what it covers. • Clear, human explanations when a model can't run or stops generating early, instead of raw technical errors. • More trustworthy leaderboards: runs that end too early or produce broken output stay on your device and never post misleading speeds.

NEW: cross-platform GGUF engine (llama.cpp). Download and benchmark GGUF models alongside Apple-native MLX, each on its own leaderboard, with GGUF results comparable cross-platforms. NEW: Manage Downloaded Models in Settings: see how much space each model uses, delete one, or clear them all. NEW: The floating monitor card is now draggable, and tapping it jumps straight to the running benchmark. NEW: Get a heads-up when a newer version of PulzeMark is available. IMPROVED: A clear "Loading model into memory" phase shows before generation starts on large models. IMPROVED: Scoring recalibrated for consistency across devices and platforms. Multi-Core scores may read lower than before; leaderboard comparisons stay fair because everyone is measured the same way. IMPROVED: Benchmarks now stop safely if the app moves to the background, protecting score accuracy. FIXED: More accurate prompt speed measurement for GGUF models. FIXED: iPhone XS, XS Max, XR and 2018-2020 iPads now show their proper names in Device Info.

AI tab redesign: your latest tokens/sec now sits right next to the neural pulse, and while a benchmark runs the screen stays focused on the live answer and system monitor. New 1024-token endurance length, plus an "Epic story" prompt built to fill long runs. The default length is now 256 tokens. The 64-token option was retired: runs that short measured startup noise, not real speed. Battery section in Device Info: charge level, power source, and Low Power Mode at a glance. Low Power Mode protection: the app now warns you before benchmarking with Low Power Mode on, and those runs stay on your device so throttled scores never skew the community leaderboards. iPhone 17e is now recognized by name. Small fixes and polish.