Every AI coding tool's free tier, compared — and what they won't tell you

A comparison of eight AI coding tools' free tiers reveals that GitHub Copilot is the only one publishing complete, hard numbers for what users get at every tier. Other tools like Cursor, Windsurf, and Replit rely on vague terms such as 'generous' or 'limited' without committing to specific quotas. The analysis, verified against official pricing pages and docs on July 5, 2026, highlights a lack of transparency across the industry, with Copilot standing out for its detailed credit system and conversion rates.

Verified against official pricing pages and docs on July 5, 2026. These numbers change without notice — if you spot a stale one, tell us and we'll fix it with a changelog entry. Every AI coding assistant has a free tier. Almost none of them will tell you, in numbers, what it contains. We pulled the official pricing pages and docs for eight tools and graded each on one axis: does the company publish hard numbers for what you get — or vibes? Scale: HARD NUMBERS / PARTIAL / VAGUE . GitHub Copilot is the only tool of the eight that publishes complete numbers at every tier: 2,000 completions a month free; $10/mo buys unlimited completions plus exactly 1,500 monthly AI credits 1,000 base + 500 flex , with a published conversion rate — one credit is one US cent — and published credit tables all the way up the tier ladder docs.github.com . You can compute what you're buying. Nobody else in this table fully lets you. Cursor publishes a real dollar number for API agent usage on Pro $20 of it per month, cursor.com — then describes the other, larger bucket as "generous Auto and Composer usage." How generous? Not a published number. The free Hobby tier is just "limited." Limited to what? Also not a number. One more thing, because stale SEO posts keep claiming otherwise: Cursor's 7-day Pro trial no longer exists. Cursor staff confirmed the removal on the official forum in January 2026 and reconfirmed in June forum.cursor.com . If a blog tells you there's a trial, check its date. Windsurf rebuilt its billing around a "quota" in March 2026 and describes it as "light" free or — you guessed it — "generous" paid . Credit where due: its own blog publishes estimated daily message ranges per model class e.g., 7–27 messages/day on top-tier models for free users, explicitly labeled estimates — devin.ai/blog , and the dual daily+weekly reset is documented, including the genuinely thoughtful detail that the daily quota exceeds 1/7 of the weekly one so weekend coders aren't punished. Hedged ranges beat adjectives. Still not commitments. Replit publishes a hard number for paid $25 of credits on Core and none at all for free: "daily credits… up to a monthly cap." Neither the daily amount nor the cap appears anywhere in the official docs docs.replit.com . v0 publishes what looks like the table's second complete answer — $5/month in free credits plus a daily message limit — until you find Vercel staff on the forums explaining the daily limit fluctuates with regional server load. A published number that isn't a promise is a new genre. Claude Code comes with Claude Pro $20/mo . Anthropic's official answer to "how much usage?" is: "at least five times the usage per session compared to our free service" support.claude.com — five times a free-tier number that is published nowhere. There's a 5-hour session reset and a weekly cap on top; no quantities for either. Google Antigravity which absorbed the old free Gemini CLI/Code Assist path gives free users "basic weekly rate limits" and AI Pro subscribers a "high, generous quota, refreshed every five hours" antigravity.google/docs/plans . Numbers: none. Upper tiers are sold as "5X" and "20X" a base that — say it with us — is not published. Consumer-facing ChatGPT pricing describes Codex access as "Limited" free and "Expanded… Limits apply" paid . Meanwhile OpenAI's developer-facing Codex pricing page publishes actual numeric per-5-hour usage ranges by plan and model. Same company, same product: adjectives for consumers, numbers for developers. Draw your own conclusion about which audience they think checks. Want to know exactly what you're buying? Copilot is the only complete answer at every tier. Best hedged-but-honest attempt: Windsurf's published estimate ranges. Wildly capable free tier, zero committed numbers: Antigravity real frontier-model access on the $0 plan — you just can't know how much . If a tool's own pricing page won't commit to a number, assume the number can change the day after you subscribe. That's not cynicism; that's what "limits may vary" means. No referral links in this post. Sources are linked inline; all quotes pulled July 5, 2026.