Ai2 Asta vs STORM

Ai2 Asta

• Researchers who want a stated reason for each paper a search returns
• Getting a cited overview of a question that spans many papers
• Exploring a tabular dataset with analysis code that stays visible
• Developers building or evaluating scientific AI agents

STORM

• Getting a cited overview of an unfamiliar topic
• The pre-writing stage of a long article or report
• Developers who want an open-source research pipeline to adapt

Ai2 Asta

• Ai2 publishes no pricing page; the app runs without an account, and its credit panel lists no credit cost for the three main agents.
• The getting started guide says Generate a report searches computer science, medicine, environmental science and biology papers.
• Data explorations stay active for 24 hours and then become read-only, and uploaded datasets are deleted after 7 days.
• Ai2 says not to upload sensitive or regulated data to the hosted version; that requires a self-hosted deployment.

STORM

• The hosted site is a research preview with limited safety measures, and its terms warn that reports can contain mistakes.
• Inputs on the hosted site are collected for research and may be distributed under a Creative Commons license.
• The developers say the output is not publication-ready and often needs a significant number of edits.
• Running the open-source package yourself requires your own language model and search API keys.

Ai2 Asta

Find papers

Breaks a query into parts, searches, follows citations and judges relevance, then lists papers with a note on why each fits. A slower deep search mode handles complex questions.

Generate a report

Answers questions that need many papers with a structured report: expandable sections, comparison tables and inline citations with excerpts. Uncited text is labeled as model-generated.

Analyze data

Upload CSV, TSV, Excel, JSON, HDF5 or Parquet data and ask questions in plain language. Asta writes and runs Python and returns findings with charts, the code and a methods section.

Follow-ups and sharing

Reports can be refined over several turns, with specific papers referenced by an @ mention. A Share button copies a link to the results of any query.

AstaLabs prototypes

Hosted previews include AutoDiscovery, which generates and tests hypotheses on a dataset, and Paper + Figure QA, which answers questions about figures and tables in papers.

Open-source agents and AstaBench

Ai2 publishes open-source versions of the agents behind the three tools, along with AstaBench, a benchmark suite and leaderboard for scientific AI agents.

STORM

Report generation

Researches a topic through Internet search, drafts an outline and writes a full-length, Wikipedia-like article with citations to its sources.

Perspective-guided questions

Looks for different perspectives on the topic and simulates a conversation between a writer and a topic expert, grounded in sources, to work out which questions to ask.

Co-STORM

A collaborative mode in which the user takes part in a discussion among AI experts, with a moderator that raises questions from information not yet used.

Mind map

Co-STORM keeps a mind map that sorts the collected information into a hierarchy of concepts, updated as the discussion goes on.

Open-source Python package

The knowledge-storm package installs with pip and is MIT-licensed. Language and embedding models are configured through litellm.

Pluggable search

Retrieval modules cover Bing, You.com, Brave, Serper, Tavily, DuckDuckGo, SearXNG and Google search, plus a vector store for grounding on your own documents.

Ai2 Asta

• Free$0

  Find papers, Generate a report and Analyze data, usable anonymously or with a free sign-in that keeps a history of queries.

STORM

• Free$0

  The hosted research preview of STORM and Co-STORM, and the MIT-licensed knowledge-storm package for running it yourself.