AI Coding Tools in 2026: Eight Compared by Job, Three We Run and Measured

Which job does each AI coding tool do?

Three jobs, and a tool that is excellent at one is usually absent at the others. Builders take prose and return a deployed app with a database; you never open the code. Editors put an agent inside VS Code and show you every change before it applies. Terminal agents take a task and return a result, in a shell, in CI or on a schedule. Pick the job first; the list inside each job is short.

The seven sticker prices we could read sit within a $10 to $30 band. The meters do not: tokens on Bolt, dollar credits on v0, checkpoints on Replit, request credits on Copilot, plan windows on Claude, on-demand API rates on Cursor past the allowance. The meter decides the bill, and every card below names it.

Which tools do developers actually use in 2026?

One survey this year is large enough to cite. JetBrains' Developer Ecosystem Survey 2026, over 15,000 professional developers in May to July, published on the JetBrains research blog in August 2026: “around 39% of professional developers worldwide were using Claude Code at work”, up from 18% in January; GitHub Copilot “experienced a decline from 29% adoption a year ago to 21% in May–July 2026”; Cursor “from 18% in January to 12% in May–July”; Codex “from just 3% in January 2026 to 16% in May–July 2026”.

Read it as usage at work, not as quality: the same report notes Claude Code's US figure at 47% and that it is “the most used AI coding tool for 31% of developers”. Adoption says which tools people reached for this year. It does not say which one finished your task, which is why the next section only counts what we ran ourselves.

What did we measure, and on which tools?

Three of the eight, because those are the three this site is built and maintained with. Everything here is from our own logs; the rest of the roundup is prices and meters.

Builders: prose in, a deployed app out

Editors with an agent inside

Agents that run outside an editor

How should you choose?

By job, then by meter, then by a run on your own task. A builder if you will not open the code; the Lovable alternatives page compares six with the same method. An editor if you want to watch; the rule files are what to write first. A terminal agent if you want to hand over or schedule; the commands and subagents guides show what that looks like in a workspace that runs on it. Anthropic's cost page is the one place a vendor publishes an average, about $13 per developer per active day for enterprise API use; use it as the scale, not the estimate.

Then run one representative task on the free tier of the two finalists and read the meter. That number beats every figure on this page, including ours.

What we deliberately did not publish

• A shared benchmark or a ranking. Same task, same repository, paid seats on all eight, token readouts from each: until it exists this page compares jobs, prices and meters, and one side's own runs.
• SWE-bench or other benchmark scores quoted by the ranking pages. They measure models on a fixed task set, not these products on your repository.
• Codex, Gemini CLI, Amazon Q, Tabnine, Cody and the rest of the twenty-tool lists. We did not read their pricing pages today, so they are absent rather than guessed.
• Windsurf as a product. Its pricing URL redirected to Devin's on 26 September 2026; the card says that and nothing more.

Sources

All checked on 26 September 2026. Prices and adoption figures are in src/data/aiCodingTools.ts.

• JetBrains research blog: AI Coding Agents: Adoption Trends (August 2026)
• Claude Code docs: Manage costs effectively
• Lovable: pricing page
• Bolt.new: pricing page
• v0: pricing page
• Cursor: pricing page
• GitHub Copilot: pricing page
• Windsurf (now under Devin): pricing page
• Claude Code: pricing page
• Replit: pricing page