Agents
Vercel's agent-desktop hits 1.0, SemiAnalysis's GDPx benchmark tops out at 12%, Perplexity rents Blackwell Ultra GPUs from IREN: the November 2 roundup
Vercel Labs ships an open-source tool that lets agents drive desktop apps. A new agent benchmark finds the best model solves 12%. Perplexity signs for GPUs.
- Coding agents
- Models
- Cloud & infrastructure
- #Vercel
- #agents
- #computer use
- #benchmarks
- #SemiAnalysis
- #Claude
- #Perplexity
- #IREN
- #NVIDIA
Three things from the weekend: a tool that gives agents a cleaner way to use desktop apps, a hard new benchmark that puts today's frontier models in their place, and one more AI company locking in GPU capacity.
Vercel Labs' agent-desktop reaches 1.0
agent-desktop, from Vercel Labs, lets an AI agent read and operate desktop apps on macOS, Windows and Linux. Instead of guessing from screenshots, it reads the operating system's accessibility tree, the structure screen readers use. A snapshot returns JSON with a reference for every element the agent can act on; the agent clicks or types on a reference, then takes a new snapshot to check the result (GitHub).
It is a command-line tool and a C library written in Rust, under the Apache 2.0 license, installed with npm install -g agent-desktop. By default it works headless: actions go through the accessibility interface without moving the pointer or taking focus. The README lists 60 commands, 56 of them usable today. On macOS it needs the Accessibility permission, and on Linux a graphical session with at-spi2-core.
Version 1.0.0 came out on October 28 and 1.0.2 on October 30 (releases). The repository had about 7,500 stars on November 2 (GitHub API). The README carries a fair warning: snapshots, screenshots and clipboard reads can contain private data, so keep them out of public logs.
GDPx: the best model solves 12% of the tasks
On November 2, SemiAnalysis (Max Kan, Andrew Megalaa and Dylan Patel) published GDPx, a benchmark for agents in 11 domains it sees as the most valuable for frontier AI: chip design, cybersecurity, biology, finance, legal, quant trading, recursive self-improvement, hardware design, materials science, alignment and software engineering. Version 1.0 has 206 tasks built with 14 specialized partners (GDPx).
The leaderboard is humbling: Claude Opus 5.5 leads with 12.0%, ahead of Gemini 4 Carbon (10.9%) and GPT-6 Astra (9.6%); GLM 5.3 closes the list at 2.7%. Each domain weighs the same and each task is scored on the average of three attempts. According to SemiAnalysis, Opus 5.5 wins quant trading, hardware design and chip design, Gemini 4 Carbon wins biology and software engineering, and GPT-6 Astra wins recursive self-improvement and materials science. On finance and legal, "basically every model scores exactly 0" (SemiAnalysis).
Two caveats from the authors themselves: safety classifiers lower the scores of Opus 5.5 and GPT-6 Astra in cyber, biology and alignment, and the cost per task varies a lot, about $51 for Opus 5.5 against about $16 for Gemini 4 Carbon (Unite.AI).
Perplexity signs for Blackwell Ultra GPUs at IREN
On October 30, IREN announced a multi-year agreement under which Perplexity will run AI workloads on Nvidia Blackwell Ultra GPUs at one of IREN's data center campuses. The capacity will serve Perplexity's research product and Perplexity Computer, which coordinates agents to take a project from one request to a finished result. IREN says more than 50,000 organizations use Perplexity (IREN), and that Perplexity users ask more than 1.5 billion questions a month (IREN on X).
IREN does not give the contract's value, length, number of GPUs or campus location.
What to keep in mind
- agent-desktop is an open-source building block, not a finished agent: the agent's loop stays yours.
- GDPx is new and run by SemiAnalysis with its partners; one benchmark is not a verdict.
- The IREN–Perplexity deal is announced by IREN only, without figures on its size.
Sources
- Vercel Labs, agent-desktop on GitHub, checked November 2, 2026
- agent-desktop releases (v1.0.0 to v1.0.2), checked November 2, 2026
- SemiAnalysis, GDPx leaderboard, checked November 2, 2026
- SemiAnalysis, "GDPx: Measuring Frontier Model Capabilities Across the Most Valuable Domains", November 2, 2026
- Unite.AI, "SemiAnalysis Launches GDPx Benchmark Spanning 11 Agentic Domains", November 2, 2026
- IREN, "Perplexity and IREN: Powering Fast, Trustworthy AI at Scale", October 30, 2026, checked November 2, 2026
- IREN, post on X, October 30, 2026
Spotted an error in this article? Report it · Our corrections policy
