Rank 71 AI harnesses from 55 makers: the coding agents, personal agents and app builders that wrap a model in tools, memory and a loop. Drag them, tap to place, or use the keyboard. Your list saves in this browser, and you can share it as a link or an image.
0 of 71 ranked
Drop harnesses here
Drop harnesses here
Drop harnesses here
Drop harnesses here
Drop harnesses here
Drop harnesses here
Drag harnesses into tiers, or tap one and then a tier. Keyboard: focus a harness, press 1–6 to rank it and 0 to
unrank, or Enter then the arrow keys to move it. Ctrl+Z undoes.
Unranked
Kind
Runs in
Source
Models
Released
Origin
Maker
Sort
Popular is the seven coding agents developers use most at work in the JetBrains Developer Ecosystem Survey 2026,
with Claude Code and Codex split into CLI and app, plus the always-on agents people run for
everything else. Switch to Current or All for 57 more.
How the harnesses score
A score belongs to a harness and a model together. Only runs on that month’s top models from
the AI Model Tier List count.
Artificial Analysis runs each harness and model pair on DeepSWE, Terminal-Bench 4.0 and SWE-Atlas QnA, three tries per task, and averages the three. Artificial Analysis, read Oct 4, 2026.
October 2026 Scored against this month’s top models.
ModelEffort only
68.4% is the top score, from Claude Code CLI with
Claude Sonnet 5.5. Each harness shows its best run on one of that month’s top models.
Claude Code CLIClaude Sonnet 5.5 · max effortClaude Code CLI with Claude Sonnet 5.5, max effort: 68.4%68.4%
Antigravity CLIGemini 4 Argon · high effortAntigravity CLI with Gemini 4 Argon, high effort: 63.8%63.8%
Codex CLIGPT-6.1 Sol · xhigh effortCodex CLI with GPT-6.1 Sol, xhigh effort: 62.9%62.9%
DevinDevin Fusion CLI · Claude Fable 5.1 + SWE-2 · xhigh effortDevin (Devin Fusion CLI) with Claude Fable 5.1 + SWE-2, xhigh effort: 61.7%61.7%