#11up 1 positions this week
Claude Code
Terminal native coding agent with the best long task retention.
934/ 1000
3,874 votes
H2H 13/4
Coding agent board
Coding agents and LLM tools, ranked into tier bands
These are the tools that work inside a repository you already have. Same ten weighted axes as the app builder board, same community voting, different question: how quickly does a correct change land.
S tier
Completes the full board brief, a real application with data, auth and a deploy target, and the output is presentable at the end.
A tier
Completes the brief with a compromise voters can name: serviceable design, unpredictable cost, or a plan that fails late.
#41down 1 positions this week
GitHub Copilot
The default, wired into the place your code already lives.
812/ 1000
2,604 votes
H2H 8/7
B tier
Completes a narrower version of the brief, or completes it in a way that does not survive several rounds of iteration.
#60No change in rank this week
Gemini CLI
Open source terminal agent with a very large context window and a free tier.
704/ 1000
1,122 votes
H2H 4/7
Full ranking
Every entry, in board order
| Rank | Tool | Tier | Score / 1000 | Votes | Trend | H2H | Verified |
|---|---|---|---|---|---|---|---|
| #1 | Claude Code | S tier | 934 | 3,874 | 1up 1 positions this week | 13W 4L | 18 Aug 2026 |
| #2 |  Cursor | S tier | 908 | 3,561 | 1down 1 positions this week | 11W 5L | 17 Aug 2026 |
| #3 |  Codex CLI | A tier | 851 | 2,288 | 2up 2 positions this week | 7W 5L | 17 Aug 2026 |
| #4 | GitHub Copilot | A tier | 812 | 2,604 | 1down 1 positions this week | 8W 7L | 16 Aug 2026 |
| #5 |  Windsurf | B tier | 736 | 1,489 | 1down 1 positions this week | 5W 8L | 14 Aug 2026 |
| #6 | Gemini CLI | B tier | 704 | 1,122 | 0No change in rank this week | 4W 7L | 15 Aug 2026 |
What qualifies for the coding agent board
An entry belongs here if its unit of work is a change to code you already own. Terminal agents, editor integrations, CLI tools and review bots all qualify. The tool does not have to host anything, provision a database or produce a user interface, and it is not penalised for leaving those to you, because on this board they are not part of the job.
How the axes read differently here
Design carries the same 22 points, but voters are rating a different artefact. On the app builder board design means the first render. Here it means the shape of the change: does the diff read like it was written by someone who had seen the rest of the codebase, or does it invent a second pattern next to the one already there. Tools that score well on this axis tend to be the ones that read before they write.
Speed carries 21 points and is measured end to end, which quietly penalises the fastest raw generators. A tool that produces a change in fifteen seconds and then needs three rounds of correction is slower than one that takes ninety seconds and lands. Community votes reflect that lived total rather than tokens per second.
Agent performance carries 11 points and is the axis with the widest spread on this board. It measures how far a tool gets on a multi step task with nobody intervening, and specifically how it behaves after it makes a mistake. The strong entries notice a failing test and change approach. The weaker ones repeat the same fix with more confidence.
Terminal against editor
The two dominant form factors trade wins predictably. Terminal agents run longer without supervision, hold more of the repository in view and handle sweeping refactors better, so they take agent performance and scalability. Editor integrations keep the human in the loop on every hunk, which makes them faster and more reliable on small, well specified edits, so they take the review loop and reliability.
If your work is mostly small changes in code you know well, the editor profile probably wins for you regardless of the overall order here. If your work is migrations, sweeps and unfamiliar subsystems, weight agent performance far higher than the 11 points it carries on the published weights and the ranking will reorder for you.
Why we keep the two boards apart
Because merging them would produce a ranking nobody could act on. Half the entries would be scored on integrations they deliberately do not ship, and the other half on repository awareness they never need. Keeping them separate means a vote on either board measures one thing, and the shared axes and weights mean you can still compare across them when the comparison is genuinely meaningful.
Questions
About this board
What is the best coding agent in 2026?
On this board Claude Code leads on community votes, with Cursor close behind on a different profile: the terminal agent wins agent performance and long task recovery, the editor wins on the tightness of the review loop. Neither margin is large enough to call the question settled.
How is a coding agent different from an AI app builder?
A coding agent works inside an existing repository and hands you a change to review. An app builder starts from nothing and hands you a running application including hosting and a data layer. They are voted on separate boards because a vote on speed means something different in each case.
Do the same weights apply to this board?
Yes, the same ten axes and the same weights, so the two boards stay comparable. Design reads differently here: it is the quality of the code and diff the agent produces rather than the visual output, because these tools do not render an interface for you.
Are terminal agents rated against IDE extensions fairly?
They are rated on the same axes, which favours whichever form factor gets a correct change merged faster. Terminal agents tend to win agent performance and long horizon tasks; editor integrations tend to win the review loop and reliability on small edits. Both patterns are visible in the axis breakdown on each tool page.