Skip to content
Agents tracked: 492 Releases (7d): 545 Agent status: 1 with issuesSince last limit reset: Claude 9d · Codex 7d For AI agents: Markdown Updated Nov 20, 2026

Autoresearch vs Horizon

Autoresearch (karpathy) and Horizon (Sheng Qiang) side by side on public data, as of Nov 20, 2026.

Autoresearch has 6.8x the GitHub stars of Horizon (100K against 14.8K).

Side-by-side numbers

Metric AutoresearchHorizon
Gid Score4041
Usage data as ofNov 19Nov 19
Rank by Gid Score#85#82
Downloads, last 7 days——
Downloads, 7-day change——
Downloads, 30-day change——
VS Code installs——
Marketplace rating——
GitHub stars100K14.8K
Homebrew installs, 30 days——
Hacker News mentions, 30 days150
Latest stable release——
Terminal-Bench (best run)——
PriceFree (OSS)Free (OSS)
LicenseMITMIT
First releasedMarch 2026February 2026
InterfacesSelf-hosted—
Models——

A dash means there is no public data for that metric as of Nov 20, 2026.

Trends

GitHub starsrunning total
AutoresearchHorizon

Autoresearch

Andrej Karpathy's experiment in letting an AI agent do ML research on its own: the agent edits a small GPT training script, trains for a fixed five minutes on one GPU, keeps a change only if validation bits per byte improve, and repeats overnight.

Pricing: Open source under the MIT licence; it needs your own NVIDIA GPU (tested on an H100) and a coding agent you point at program.md.

Full Autoresearch profile

Horizon

Self-hosted AI news radar: it collects items from your feeds and forums, filters and deduplicates them, and delivers a ranked daily briefing in English or Chinese with context and community discussion, by email, Feishu or WeChat.

Pricing: Open source under the MIT licence; check the website for any paid hosted plans.

Full Horizon profile