The AI Free Tier Reality Check: What 15 Tools Actually Give You (Documented Limits, 2026)
TL;DR: We read the official pricing and help pages of 15 popular AI tools to document exactly what their free tier gives you. The pattern is clear: the big chat tools (ChatGPT, Claude, Perplexity) publish no numeric limit at all — it’s deliberately dynamic. The “free forever” credit tools mostly reset daily and don’t roll over (Leonardo 150/day, Suno 50/day, Lovable 5/day) or are one-time only (Runway’s 125 credits). And the real catch is usually the watermark or the no-commercial-use clause, not the quantity.
Read this first — our honesty rule. Everything below is the vendor’s documented limit, quoted from their own pricing or help pages (dated). We did not sign up and hands-test these limits, because documented limits are publicly verifiable and hands-testing 15 accounts isn’t. Where a vendor’s pages were blocked and we couldn’t confirm, we say UNVERIFIED rather than guess. Vendors change these constantly — if you spot something out of date, email us at [email protected] and we’ll fix it with credit.
Every AI landing page has a big “Start Free” button. Very few tell you, on that same button, what “free” actually costs you in limits. So we did the reading for you.
The three kinds of “free”
After documenting 15 tools, free tiers fall into three honesty buckets:
- The undisclosed dynamic cap — no number published; you just hit a wall and wait. (ChatGPT, Claude, Perplexity)
- The resetting credit bucket — a small daily or monthly credit grant that does not roll over. (Leonardo, Suno, Lovable, ElevenLabs, HeyGen, Descript, Otter, Grammarly)
- The one-time taster — a fixed grant that never refreshes, or a straight trial. (Runway’s 125 credits, Notion AI’s trial)
Group 1: The chat tools that won’t tell you the number
This is the most important finding. The three most-used AI assistants do not publish a numeric free-tier limit.
| Tool | Documented free limit | The catch |
|---|---|---|
| ChatGPT | “A limited number” of latest-model messages “within a five-hour window” — no fixed number | Rolling 5-hr cap you can’t see; separate stricter limits on image/file/data tools; can’t build GPTs; ads rolling out in some regions |
| Claude | Dynamic — limits “vary by conversation length, complexity, model, and effort” — no fixed number | You get cut off mid-task and wait for a reset window; 200K context on free models |
| Perplexity | “$0/month — good for limited daily usage”; “limited weekly usage” of top models | No numeric Pro-search count published; no premium sources, no preferred model |
Why this matters: you literally cannot budget your usage, because there’s no published unit. The limit is whatever their capacity planning decides today. That’s a legitimate engineering choice — but it’s the opposite of pricing transparency, and it’s worth knowing before you build a workflow on a free plan that can silently shrink.
Sources: OpenAI Help — ChatGPT Free FAQ, Anthropic — usage limits, Perplexity pricing. Confidence: medium — verified that no fixed number is documented.
Group 2: The credit buckets (and how fast they reset)
These publish a real number — but read the reset cadence carefully. “1,800 credits a year” sounds better than “5 a day,” but they’re the same thing, and the daily version means you can’t save up.
| Tool | Documented free limit | Reset cadence | The real catch |
|---|---|---|---|
| Leonardo AI | 150 tokens/day | Daily, no rollover | A single premium (Alchemy) image can eat 8–16 tokens |
| Suno AI | 50 credits/day (~10 songs) | Daily, no carryover | No commercial use; shared/slow queue; no stem separation |
| ElevenLabs | 10,000 credits/mo (~10 min TTS) | Monthly, no rollover | No commercial license on free (starts at $6 Starter) |
| HeyGen | 3 videos/mo, ≤1 min, ≤1080p | Monthly | Watermarked; standard (slow) processing |
| Descript | 60 min transcription/mo | Monthly | Watermark-free export is a paid upgrade |
| Otter.ai | 300 min/mo + 3 lifetime file imports | Monthly minutes; imports are lifetime | You get only 3 uploaded-file transcriptions ever |
| Grammarly | 100 AI prompts/mo | Monthly | No full rewrites, tone adjustment, or detectors |
| Lovable | 5 build credits/day (~30/mo) | Daily, expire same day | Credits vanish at day’s end; no persistent balance |
Two traps to internalize:
- “No rollover” is the norm, not the exception. Almost every free credit grant here resets and forfeits unused credits. If you only create on weekends, a daily-reset tool wastes ~5/7 of your allowance.
- The watermark / commercial-use clause is the actual paywall. ElevenLabs and Suno give generous free quantity but forbid commercial use. HeyGen and Descript watermark free output. For anything client-facing, the free tier is a demo, not a tool.
Sources: official pricing pages for Leonardo, Suno, ElevenLabs, HeyGen, Descript, Otter, Grammarly, Lovable. Confidence: high, except Descript/Runway watermark which is inferred from “watermark-free” being sold as an upgrade.
Group 3: The one-time tasters
| Tool | Documented free limit | The catch |
|---|---|---|
| Runway | 125 credits, one-time (do not expire, do not refresh) + 3 projects, 5GB | Once you burn 125 credits, free is over forever; free output watermarked |
| Notion AI | A “trial of Notion AI” — “a limited number of complimentary AI responses” | Not ongoing; full Notion AI needs Business ($20/member/mo) |
Runway’s model is honest but easy to misread: “Free forever” refers to account access, not the credits. The 125 credits are a one-time welcome grant — a few dozen short generations, then you’re capped until you pay.
The two we couldn’t verify (and why that’s a finding)
| Tool | Status | Why |
|---|---|---|
| Canva | UNVERIFIED | Pricing/help pages are bot-gated; no readable official source or archive this pass |
| Midjourney | UNVERIFIED (policy: no free tier) | Docs bot-gated; its known policy since 2023 is subscription-only, but we couldn’t re-confirm from an official page today |
We’re listing these as unverified rather than repeating widely-quoted numbers we couldn’t stand behind. That’s the whole point of this audit: if we can’t confirm it from the source, we don’t print a number.
How to actually evaluate a free tier
- Find the reset cadence, not just the number. “Daily, no rollover” is the single most-hidden detail. Multiply daily by ~30 only if you’ll actually use it every day.
- Search the page for “watermark” and “commercial.” These two words decide whether the free tier is usable for real work.
- If there’s no published number, assume it’s small and shrinking. Dynamic caps favor the vendor’s costs, not your workflow.
- Screenshot the free-tier terms the day you sign up. Vendors quietly cut free allowances (we’ve tracked this — see our AI price-change tracker).
We publish our full scoring method and per-tool trust data separately: how we score tools · the Trust Index.
Method: documented limits read from official vendor pricing/help pages, 2026-07 (several JS-gated pages read via July 2026 Wayback snapshots of the official URL). Nothing hands-tested. Corrections welcome at [email protected].