Cortex Discover 1.1.0: Pervaziv's agentic AI browser uses 36.7% less memory than Chrome, six-browser study finds

Pervaziv AI published benchmark results for Cortex Discover 1.1.0, its dedicated agentic AI browser, ranking it first in memory and CPU efficiency across Chrome, Safari, Brave, Edge and Firefox — with throughput and rendering near-identical to Chrome. The company also disclosed where the browser still trails.

Pervaziv AI is making an unusual pitch for its agentic AI browser: not more capability, but less appetite. The company on October 6 published performance results for Cortex Discover 1.1.0, its dedicated browser for AI-assisted enterprise work, claiming it beat five mainstream browsers on resource efficiency while matching Chrome on the things users actually feel.

The numbers come from a structured, company-run study that put Discover alongside Google Chrome, Safari, Brave, Microsoft Edge and Firefox. Every browser ran the same repeatable workload — startup, memory, CPU, navigation, throughput, rendering, responsiveness, background activity and synthetic video — under the same test conditions, with 36 measured runs completing successfully and Chrome serving as the primary reference.

The headline numbers

Discover ranked first of the six in both the memory-efficiency and CPU-efficiency categories. Against Chrome specifically, the company reports:

• 36.7% less memory with one idle tab (439.5 MiB vs. 694.3 MiB)
• 19.9% less with multiple idle tabs (1,173.6 vs. 1,465.8 MiB)
• 10.9% less during active work (1,242.4 vs. 1,394.4 MiB)
• 19.3% faster warm startup to page-ready (915.2 ms vs. 1,134.2 ms), and 20.2% faster native launch
• 17.6% less aggregate navigation wall time (68.2 ms vs. 82.8 ms)

Crucially, the resource savings did not come from doing less. Both browsers delivered roughly 60 operations per second of throughput and rendered at roughly 60 frames per second in the measured workload. During synthetic video, both recorded zero dropped frames — while Discover used 47.4% less captured memory (623.8 vs. 1,186.1 MiB). CPU followed the same pattern: 2.4% of one core vs. Chrome's 4.6% at one idle tab, 34.0% vs. 34.8% during the active workload, and 32.7% vs. 38.2% while browsing.

Why efficiency is the pitch

The logic is straightforward: an agentic browser that continuously expands its memory and CPU footprint undermines the productivity it promises, because the browser competes for the same resources as the dashboards, IDEs, video calls and local applications running beside it. Pervaziv's position is that the browser — where users experience AI directly — is exactly where that overhead gets felt first, in tabs competing for memory, sluggish apps and rising battery use.

Discover is built around that constraint. It combines familiar browsing with native page awareness, selected-tab reasoning, visual context, controlled browser actions and file-aware workflows, keeping conversation, selected context, attachments and task status beside the active page. With the release, the Cortex AI Model Ensemble grew from 48 to 50 specialized agents, adding browser-focused intelligence to a system that coordinates models, search, skills, safety, security, coding, privacy and verification. The first release supports Windows x64 and macOS on Apple silicon and Intel, while Pervaziv continues to ship Cortex extensions for the five mainstream browsers.

The gaps, published too

Unusually for a vendor benchmark, the release discloses where Discover loses. Handler-to-frame p95 measured 72.3 ms against Chrome's 65.5 ms — 10.5% slower — and background tab readiness p95 was 79.1 ms vs. 61.3 ms. Pervaziv says a separate same-version validation around a focused optimization cut background tab readiness p95 by 18.9% and long task time by 12.7%, and frames the misses as an engineering roadmap rather than numbers to bury.

Part of a longer performance program

The Discover results extend a performance campaign Pervaziv has run through Cortex during 2026: an April frontend/backend optimization drive that cut average response time from roughly 530 ms to 82 ms across five operations; Cortex 4.0's reported 2.5x faster coding tasks; September's semantic context compaction, which reduced eligible supplemental context by 46% while preserving recall and task success; and a three-tier inference cache architecture the company says delivered up to 150x faster prompt prefill in evaluated paths. The measurements come from different workloads and should not be combined into a single speed claim, but the direction is consistent: cut repeated work wherever it occurs.

As always with vendor-run benchmarks, independent verification would strengthen the claims — Pervaziv's own caveat is that page-level metrics moved in both directions and the aggregates are evidence of a foundation, not a sweep. But the framing is notable: in a market where AI browsers compete on how much they can do, Pervaziv is betting the differentiator is how little of your machine they take to do it.

Sources: Pervaziv AI via EIN Presswire, October 6, 2026; World Wealth Journal republish.