Models and plans

ResearchAnything has five research models. They share the same search and reading tools and differ in how much work they do and what they check. Your plan decides which models you can use and how many runs you get each month.

01The models

Pick a model from the model chip in the research box. Hover a name to see its one-line description.

RA-Lite

RA-Lite plans a few search angles, then lets the model steer the search round by round: it writes queries, looks at what came back, chooses which pages to open, and decides whether it has enough. It reads the strongest sources closely and writes the answer from them. It has no depth setting.

RA-1

RA-1 runs a deeper version of the same process, and learns across runs:

• Recalling strong sources. When a topic has history, sources that did well on it in earlier runs are brought forward before the first search.
• Credibility that learns. A domain's score blends fixed signals with its record across runs: how often its claims held up, how relevant it has been to the topic, and how reliably its pages could be read.
• Evidence and contradictions. RA-1 pulls evidence from each source, says where sources disagree, and tries to settle disagreements that earlier runs on the topic left open.
• A critic. On Deep Dive and Exhaustive, a critic reviews the draft before it ships. Exhaustive also revises it.

RA-2

RA-2 does RA-1's research, then checks every factual sentence and table row of its answer against the passages it read from the source that sentence cites. What fails is fixed before you read it:

• a marker that points at no source is removed;
• a citation that points at the wrong source moves to the source that backs the sentence;
• a supported sentence left uncited gets its citation;
• citations that do not back their sentence are dropped;
• a sentence no source backs is rewritten from the nearest evidence, and kept only if it then passes. Otherwise it is taken out and listed under the answer's uncertainties.

RA-2 also drafts under stricter rules: a citation on every factual sentence, figures copied from the source rather than recalled, and gaps stated plainly. A run on RA-2 costs one research run, the same as RA-1. The release post, with the measured comparison against RA-1, is at researchanything.ai/models/ra2.

Max

Max starts with a full RA-1 run. Then it asks what the sources so far do not cover, writes new searches for those gaps, reads the results, and keeps any source that beats the weakest one it already has. It does this for up to two more passes, and stops early when a pass adds little. When new sources are admitted, the answer is revised so they can change it. On Pro, Max runs come from a separate monthly allowance; see How usage counts.

Auto

Auto reads the question and picks one of three strategies, then runs RA-1's pipeline with it:

• Empirical — one specific, testable question, answered from primary studies. It searches the research literature by study design and reads the primary work.
• Landscape — a topic across a span of time. It searches each period and traces what changed.
• Current — a question answered by the live public record: recent events, companies, products, policy as it stands. It checks the web and separates original reporting from repeats.

The run says which strategy it chose when it starts. If you set Sources to Web or Papers yourself, your choice wins over the strategy's. When the shape of the question is unclear, Auto falls back to simple rules based on its wording. A run on Auto costs one research run and does not use a Max run.

02Depth modes

RA-1, RA-2, Max and Auto take a depth setting, under "How deep" in the model menu.

03Where to look

The Sources chip sets where a run gathers evidence. It is available on every plan and every model.

More on how sources are found and graded is in Sources and library.

04Plans

Every plan includes projects, run history, Cite, citations and exports. Researcher adds early access to beta features and priority support. Current prices are on researchanything.ai/pricing.

05How usage counts

• Monthly. Runs and chat messages are a calendar-month allowance that resets on the 1st.
• One question, one run. Each research run uses one run, whichever model answers it. RA-2's citation check is included.
• Max is part of the total. A run on Max uses one research run and one Max run. On Pro, 10 of the 40 runs can be Max runs. On Researcher, any run can be.
• Runs from chat count. A research run the project chat starts for you uses one research run.
• Chat messages. Each message you send in a project chat counts against that month's chat allowance.
• Saved sources are a cap, not a meter. The limit is on how many distinct sources your account keeps. Saving a link you already have does not use another slot.
• Cite is not metered. Citing links and running Cite on your writing do not use research runs.
• No overage. At a limit, new runs or messages wait until the reset or an upgrade. Nothing is billed automatically.

Profile shows your runs this month, Max runs where your plan has a separate count, chat messages, your saved-source cap, and how long until the reset.

06Upgrading

There is no self-serve checkout yet. To move to Pro or Researcher, create a free account, then email hello@researchanything.ai from the address you signed up with. The new allowance applies as soon as your account is moved. You can switch plans, including back to Free, the same way. Plans are per account; there are no seats.