AI coding assistants, compared on data — and one shared benchmark
Eight AI coding assistants, profiled from their vendors' own pricing pages: what the free tier really includes, what the paid tiers cost, and which form each tool takes. On top of that data layer we are running every free-tier tool through the same frozen benchmark repo, scored by unit tests — and until a run is complete, its pages say so. No invented scores, no borrowed screenshots.
The eight tools
GitHub Copilot
AI pair programming as an add-on for the editor you already use.
Free: Yes — 2,000 completions/mo + AI Credits allowance
Free tier · test scheduledCursor
An AI-first editor — a VS Code fork — built around agent workflows.
Free: Yes — limited Agent/Chat requests (limits not published)
Free tier · test scheduledWindsurf (Devin Desktop)
The editor formerly known as Windsurf — renamed Devin Desktop by Cognition in June 2026.
Free: Yes — unlimited completions + a lightweight agent allowance
Free tier · test scheduledKiro
AWS's spec-driven agentic IDE — you write the spec, the agent builds to it.
Free: Yes — 50 credits/mo, incl. Claude Sonnet 4.5
Free tier · test scheduledOpenAI Codex
OpenAI's coding agent, bundled into ChatGPT plans instead of sold separately.
Free: Yes — included with ChatGPT Free (limits not published)
Free tier · test scheduledReplit (Agent)
The zero-install option: browser IDE, agent and hosting in one web app.
Free: Yes — daily Agent credits; publish 1 app
Data profile onlyClaude Code
Anthropic's terminal-first coding agent, bundled with Claude subscriptions.
Free: No free tier
Data profile onlyTabnine
The privacy-first assistant, built for private deployment inside your own walls.
Free: No free tier (14-day trial, no card)
Head-to-head comparisons
Cursor vs GitHub Copilot
Pricing, free tiers and a data-based verdict — benchmark block included, scores pending.
ComparisonWindsurf (Devin Desktop) vs Cursor
Pricing, free tiers and a data-based verdict — benchmark block included, scores pending.
ComparisonKiro vs GitHub Copilot
Pricing, free tiers and a data-based verdict — benchmark block included, scores pending.
Why this site exists: 2026 broke every comparison you have read
This category repriced and renamed itself in a single year. GitHub Copilot moved to credit-based metering in June 2026. Windsurf became Devin Desktop the same month. Cursor closed its old Pro tiers to new users in October 2026 and replaced them with a new line-up. Amazon Q Developer was succeeded by Kiro. Any comparison written before those changes — which is most of them — is now citing retired prices, old names, or both.
So this site treats pricing data as perishable: every profile names its source and the date it was read, and the data layer is re-checked monthly. When a vendor changes a plan, the profile changes — and the read date moves with it.
How a comparison gets made here
Three layers. First, the data layer: pricing, free-tier terms and platform facts from official vendor pages. Second, the hands-on layer: one frozen benchmark repo, the same six tasks for every tool, default settings, free tier, scored by unit tests, with our own screenshots. Third, the verdict layer: a written conclusion for each pairing that cites the first two layers — and says plainly which layer it is standing on. The full method is on the How We Test page.
Current status: profiles and data comparisons are live; hands-on runs for the six free-tier tools are scheduled, and Claude Code and Tabnine are data-only because neither offers a standing free tier.