Three "size" axes (credits, chat-history retention, AI conversation memory) and four "feature" axes (chat AI quality, word detail cards, chat tone options, AI verbosity profile). Every other feature — all 33 languages, every accent, phoneme-level scoring, word-lookup tooltips, translate, progress tracking — is identical across all four tiers. See the cards above plus the "Every plan includes" section for the full list.

Both are real, production AI models. Starter uses a fast lightweight model that's perfectly fine for basic feedback. Pro and above route chat through a more capable model that's noticeably better at nuanced corrections, idiomatic phrasing, and tracking long conversation threads. The model class is an implementation detail we may swap as better options ship — we don't guarantee a specific vendor, just the tier of capability.

Most language apps teach textbook formal speech. Real people don't talk like that. AnyAccent lets you choose how casual or formal the AI plays back: "Neutral" is textbook (great for early learners), "Casual" is everyday spoken, "Sitcom" is TV-show style (e.g., Friends, La que se avecina), and "Streetwise" is the most informal register (English only). Each tier unlocks one more level. The chat register knob is admin-tunable; with subscription, you choose your own from Settings.

It's one knob that controls two things together: how long the AI's chat replies are, and how deep its correction explanations go. Four levels: Concise (short replies + brief corrections, best for fast practice), Standard (full-sentence replies + rule-level corrections), Expanded (longer replies with examples + detailed corrections), and Coach (the AI talks back like a real tutor + teacher-style multi-paragraph corrections with alternatives). Each tier unlocks one more level. Coach mode is Ultra-only and opt-in from Settings — it's not the default because longer AI output costs more credits per turn; we let you turn it on when you want it.

Credits meter usage so the price matches what AI compute actually costs us. Practice is a fixed 5 credits per attempt. Chat costs 4-38 credits per turn depending on the reply length and whether corrections are on (and how deep they go). Word detail cards are 3-20 credits depending on the depth you pick. Word lookup tooltips, translations, and native voice playback are always free. The catalog endpoint shows the exact cost table so the UI can render ranges instead of a single estimate that misleads.

AI chat data adds up fast — a daily learner generates thousands of stored messages per month. Capping retention by tier keeps the app fast, your storage footprint small, and our infrastructure cost honest. If you need a transcript longer than your tier's window, export it from the chat page before it rolls off.

AnyAccent covers 33 languages, from Mandarin and Japanese to Tamil, Catalan, and Arabic. Each language uses its native locale model — not an English approximation. See the full list on our Languages page.

Yes — that's the core idea. AnyAccent treats accent as a setting, not a separate course. Switch from US to UK English (or São Paulo to Lisbon Portuguese) anytime from the language picker. Every tier has every accent.

Within 7 days, if you've completed fewer than 30 practice sessions, you can refund yourself in one click from Settings → Billing. Above that threshold, email support@huginara.com and we'll review the request case by case. Full policy in our Terms.

We don't offer a free trial. The 7-day self-serve refund above is the equivalent of trying without commitment.

Yes, anytime from Settings → Manage subscription. Upgrades take effect immediately and we prorate the difference. Downgrades take effect at your next renewal — you keep the higher tier through what you have already paid for.

Anytime from Settings → Manage subscription. Your subscription stops renewing and you will not be charged again. You keep access until the end of the period you've already paid for. Cancelling doesn't issue a refund — see the refund question above if you want money back.