NVIDIA HORIZON: Autonomous Agent Completes 100% of RTL Benchmarks in Chip Design

In July 2026, NVIDIA unveiled HORIZON, an autonomous agent for designing digital circuits at the RTL level. The company says it is the first in the industry to complete 100% of tasks on standard RTL benchmarks.

What RTL is and why it is hard

RTL (Register Transfer Level) is a low-level description of chip logic written in specialized languages: Verilog, VHDL or SystemVerilog. At this level, engineers describe how data moves between the registers of a processor or GPU on every clock cycle. Writing correct RTL is slow, high-stakes work: a single logic error becomes a hardware bug that cannot be fixed with a software update after manufacturing.

Before HORIZON, automated systems handled RTL tasks only partially: tools suggested code fragments but could not iterate on their own and carry a task through to a complete working solution.

How HORIZON works without human involvement

The agent's architecture rests on isolating each RTL task in its own git repository using git worktrees. The agent acts like an experienced engineer: it writes code, runs verification, analyzes errors and makes fixes until the task is solved correctly. No operator intervention is required at any step.

• Full autonomy: there is no point at which a human is needed
• Isolation: each task lives in its own versioned repository
• Iterative verification: the agent fixes the errors it finds on its own
• Mechanism: git worktrees for parallel, independent exploration of options
• Result: 100% completion of tasks on standard RTL benchmarks

This approach lets the agent keep a full history of changes, roll back to working states and, when needed, pursue several solution hypotheses at once.

Why 100% on RTL benchmarks matters

Passing every RTL benchmark is a significant milestone for hardware design automation. Such tasks previously required specialists with years of experience in digital circuit synthesis. For NVIDIA, which designs some of the most complex GPUs in the world, a tool like this opens a path to automating a large part of the development cycle for future generations of accelerators.

If the results hold up on real commercial projects, HORIZON could shift the balance between hand-written and automated RTL: engineers could move from routine circuit coding to high-level architectural work.

What this means

Autonomous agents are moving into hardware design, one of the most technically demanding engineering disciplines. NVIDIA's HORIZON result suggests that AI systems can already close full RTL task cycles without human involvement. The next question is how well the agent scales from benchmark tasks to real industrial projects on the level of NVIDIA's future architectures.

Frequently asked questions

What are RTL benchmarks?

RTL benchmarks are sets of standardized tasks for describing digital circuits in Verilog or VHDL. They test whether an automated system can generate syntactically correct and logically valid hardware code without human involvement.

How does HORIZON differ from ordinary AI coding assistants?

Tools such as GitHub Copilot suggest code fragments but do not manage the full cycle of solving a task. HORIZON is an autonomous agent: it launches verification itself, interprets errors and iterates until it reaches a working result, keeping the entire work history in isolated git worktrees.