Flash and Thinking

DecksMD Pro gives you two tiers to pick from when you generate or refactor cards. They are not “cheap” and “good” — both produce usable cards. They differ in how much the model works through the material before it writes anything.

Flash

The default, and the right choice most of the time.

Reach for it when the material is already clear and the job is mainly turning it into cards: lecture notes, a chapter you have read, a glossary, definitions, vocabulary, dates, a set of formulas you want to drill. It answers quickly, so you can generate, skim the inbox, keep what is good and move on.

Thinking

The model reasons through the material first, then writes. That takes longer — noticeably so on a long source — and it is worth the wait when the cards depend on getting something right rather than getting it down:

• material you have not understood yet, where a card that quietly restates a misunderstanding is worse than no card
• problems with steps — derivations, proofs, case law, differential diagnoses — where the card has to capture the reasoning, not just the answer
• a long or messy source you want broken into genuinely atomic cards rather than paragraphs with a question mark on the front
• refactoring a card you keep failing, when you suspect the card itself is the problem

If a batch of Flash cards comes back shallow or subtly wrong, that is the signal to run the same source through Thinking.

Which model actually runs

You do not pick one, and this is deliberate.

Both tiers are pointed at models we test and change as better ones appear. New models arrive constantly; a few each month are meaningfully better than what came before, and picking well means tracking price, capability and availability across hundreds of options. That is a chore, and it is the part of “use AI for your flashcards” that goes stale fastest.

So DecksMD keeps up with it for you. Each tier is measured against the current field, and when something clearly better shows up, it becomes what the tier runs. You do not change a setting, you do not re-enter a key, and you do not pay more when a stronger model lands.

If you would rather choose the model yourself, you can: add your own API key in Settings → AI and pick any model from OpenAI, Anthropic, Google, or any OpenAI-compatible endpoint including a local one. See AI assistance.

Your monthly allowance

Both tiers draw on the same monthly allowance, and Thinking uses more of it per card because the reasoning itself is work the model has to do. There is also a daily ceiling, so one runaway afternoon cannot spend the whole month.

For ordinary study this is not something you will notice. If you are converting an entire textbook in one sitting, use Flash for the bulk of it and Thinking for the chapters that actually need it — that is the split the two tiers exist for.

Your remaining allowance is shown in Settings → AI, and on your account page at decksmd.app.