AI Research Tools

Overview

AI research tools split into four buyer lanes now: literature review, academic search, citation context, and source-grounded analysis. The right purchase depends on whether the user needs to find papers, evaluate citations, analyze a fixed source set, or research current web material outside academic databases.

As of June 24, 2026, Elicit is AiPedia’s best research-workflow pick because it is built around paper discovery, screening, extraction, evidence tables, reports, API access, and systematic-review workflow; its Basic, Plus, Pro, Scale, and Enterprise ladder still needs billing-cadence verification at purchase time because research-team packaging can move quickly. Semantic Scholar is the free academic-search/API baseline, with its homepage showing 234,531,320 searchable papers and its API page still listing 214 million papers, 2.49 billion citations, and 79 million authors. Scite is the specialist pick when citation context matters, with Basic still verified at $20/month, Pro at $50/month, Organization/Developer on custom terms, and Scite MCP relevant for teams grounding ChatGPT, Claude, Copilot, Cursor, Claude Code, or other MCP clients in Smart Citations. The June 6 Scite vs Semantic Scholar refresh now makes the workflow boundary explicit: use Semantic Scholar to find and map papers for free, then use Scite when a claim, citation, or bibliography needs Supporting/Contrasting/Mentioning context. NotebookLM is best when answers must stay grounded in a reviewed source library and the buyer wants reports, Deep Research, Audio/Video Overviews, data tables, infographics, slide decks, and study artifacts with source review. Perplexity is the best cited-web add-on for current sources, documentation, policy, market, company research, and Search/Sonar API workflows. Consensus is the academic Q&A lane, with Free for testing, Pro at $15/month or $120/year for routine cited answers, Deep at $65/month or $540/year for 200 Deep reviews/month, and Consensus 2.0 Deep Search now usable inside Library and Collections for custom literature-review outputs. The Consensus pricing guide is the plan-decision route for students, researchers, clinicians, and labs choosing between Free, Pro, Deep, Teams, and Enterprise. Connected Papers is now rechecked around higher paid prices: Free is still 5 graphs/month after sign-up, Academic is USD 72/year, Business is USD 240/year, and multi-origin graphs make it more than a single-seed map when the buyer needs to refine a research neighborhood. ChatPDF is now the quick document-chat and study-tool lane, with multi-file folders, non-PDF file support, YouTube chat, AI Research, flashcards, slides, mobile apps, desktop app, and a simple API, while Humata is the paid PDF/team-document lane with Free 60 pages/month, Expert $9.99/month for 500 pages and up to 3 users, Team $49/user/month for 5,000 pages and up to 10 users, OCR, folder permissions, and SOC-2.

The June 24 nanochat refresh keeps it in the LLM check shows 55,379 stars, 7,613 forks, an MIT license, and a May 5, 2026 push timestamp, while the README still frames nanochat as a readable single-node harness for tokenization, pretraining, SFT, RLHF, evaluation, inference infrastructure.

The June 5 Elicit comparison pass refreshed Elicit vs Perplexity, and . The decision split is now cleaner: Elicit is the evidence-workflow layer, nanochat is LLM training education, Perplexity is current cited web research, Scite is citation-context verification, and Semantic Scholar is free discovery plus scholarly metadata/API infrastructure.

The June 4 ChatGPT vs Scite refresh keeps the buyer split explicit: ChatGPT is the broad research assistant for drafting, explaining, and planning, while Scite is the citation-context layer for checking whether later literature supports, contrasts, or merely mentions a claim.

The June 4 Claude vs Elicit refresh adds the matching research-buyer boundary: Claude is the better writing, critique, and synthesis assistant, while Elicit is the dedicated literature-review workflow for paper discovery, screening, extraction tables, and systematic-review-style evidence work.

The June 27 cleanup rechecked the direct research comparison lane after retiring adjacent research-discovery vs pages where field mapping, academic Q&A, systematic review workflow, citation context, and academic search were not direct substitutes. Keep Consensus vs Elicit as the direct workflow comparison, and use this hub for the broader lane split: Connected Papers maps a field, Consensus answers focused academic questions, Elicit manages systematic-review workflow, Scite checks citation context, Semantic Scholar is the free academic-search/API baseline, and nanochat is an educational LLM training harness rather than a paper-search product.

For budget-conscious research teams running their own pipelines, Reka Edge is now a physical-AI and multimodal pricing at $0.10 per million input tokens, with Edge image input at $0.005 and video at $0.03/minute. Its weights are source-available under BSL 1.1 with a commercial threshold, so buyers should not describe it as permissive Apache/MIT open source. Teams evaluating open-weight, long-context extraction or agentic coding around research corpora can also shortlist GLM: GLM-5.1 has MIT Hugging Face weights, 200K context, 128K maximum output, and public Z.AI API pricing, but it is model infrastructure rather than a literature-review workflow. The June 21 MiniMax recheck adds a similar model-infrastructure lane: MiniMax-M3 is worth benchmarking for long-context, native multimodal, coding, or agentic extraction experiments, but it is not a research workflow product and buyers must verify whether their account has >512K input or Priority access before relying on the 1M-context positioning.

The June 10 Yi refresh keeps 01.AI in this hub only as a frozen-model and enterprise-platform research edge case. Yi-Lightning and Yi open models still appear on 01.AI, Hugging Face, and GitHub surfaces, but 01.AI’s active 2026 story is WorldWise/WanZhi 2.5, multi-agent enterprise deployment, and Super Employee products. Use Yi for reproducible baselines, legacy bilingual experiments, or studying 01.AI’s pivot; use active model families such as Qwen, DeepSeek, GLM, Kimi, Reka, or MiniMax for new model-infrastructure evaluation. The June 15 Qwen recheck keeps an important research caveat in place: Qwen Cloud’s newest changelog item is still the June 8 qwen3.7-max snapshot, but the live qwen3.7-max marketplace page still describes text input/output, so modality-sensitive extraction or screen-understanding research should verify the exact endpoint.

The May 31 update adds an important caveat for high-stakes science. OpenAI’s Rosalind Biodefense expansion shows that specialist research models can be powerful but gated. GPT-Rosalind is not a general research subscription; it is a trusted-access life-sciences model program for qualified teams and public-health or biodefense workflows. Treat that as a separate procurement lane from everyday literature review.

For legal and professional-services research, Harvey belongs in a separate enterprise lane from academic tools. Its June 23 refresh keeps Assistant, Vault, Knowledge, Agents, Contract Intelligence, Command Center, Harvey Mobile, the Claude for Legal connector, and the June contract workflow update in scope. It is for matter-grounded legal research, drafting, contract review, and deployment governance inside large firms, not for student literature review. Spellbook sits in the narrower contract-drafting/review lane: the June 9 check keeps pricing custom, the trial at 7 days, and the suite focused on Microsoft Word plus Associate multi-document workflows, not broad academic search.

The June 26 interview-prep guide refresh adds a practical source-grounded research use case: use NotebookLM for a resume/job-description/source packet, Perplexity or Gemini for current company research, and a general assistant only after primary sources are opened. That keeps interview research in the low-stakes preparation lane rather than pretending a chatbot can own evidence judgment.

The June 27 medical-research archive recheck keeps that route noindexed. Current FDA, NIH, and OpenAI GPT-Rosalind/Rosalind Biodefense sources make the editorial bar higher, not lower: a live medical-research guide needs medical-specific evidence standards, privacy/PHI/IRB caveats, regulator context, clear separation from clinical decision support, and specialist life-sciences model coverage before AiPedia should rank tools for that intent.

The wrong move is treating a general chatbot as a research protocol. ChatGPT and Claude can help read, explain, outline, and critique material, but they do not replace database selection, inclusion criteria, source inspection, citation verification, or domain expertise.

Best Picks

What To Buy First

Researchers doing evidence reviews should start with Elicit plus Semantic Scholar. That covers structured workflow and free academic discovery.

Students and analysts working from a fixed set of class readings, PDFs, interviews, or reports should start with NotebookLM. It is safer for source-grounded study than asking a general chatbot to roam.

Small teams that need a dedicated PDF chat workspace should test Humata against their actual scans, tables, contracts, papers, and page volume before paying. Its Team plan is more about OCR, folders, permissions, and SOC-2 than casual single-document Q&A.

Teams checking whether evidence is supported or disputed should evaluate Scite. It is a paid specialist for citation context, not a general writing assistant.

People doing current business, policy, product, or technical research should add Perplexity. It is stronger for fast cited web trails than academic-only tools.

Large law firms and legal departments should evaluate Harvey separately from academic research tools when the work is privileged matter analysis, contract review, drafting, or firm-wide legal AI deployment.

Money Guides

• Best AI Tools for Researchers is the June 27 verified research buyer guide for Elicit, Semantic Scholar, Perplexity, Claude, NotebookLM, Scite, and Consensus across literature review, free academic search, current cited web research, close reading, source notebooks, citation context, and research-integrity rules.
• Best AI for Academic Writing is the June 26 verified adjacent guide for source-grounded writing, literature-review workflows, citation discovery, and academic-integrity-safe AI use.
• Best AI for Citations and References is the June 28 verified citation-integrity guide for Scite, Semantic Scholar, Elicit, Perplexity, NotebookLM, and Consensus, with mobile-friendly buyer paths for citation context, fixed-source notebooks, cited web research, and literature-review workflows.
• Consensus Pricing for Students and Researchers is the Consensus plan-decision guide for Free, Pro, Deep, Teams, and Enterprise, with upgrade guidance around Pro messages, Deep reviews, Library and Collections, and formal-review alternatives.
• Best AI Tools for Students is the June 27 verified student research and study guide for ChatGPT, NotebookLM, Perplexity, Claude, Cursor, Gemini, and Semantic Scholar, with source-grounded class-material study, cited-web research, academic integrity, and coding-assistance boundaries foregrounded.
• Best AI for Data Analysis is the June 26 verified adjacent guide for ChatGPT file analysis, Gemini/Google Sheets workflows, Claude analytical critique, Hex governed data-team notebooks, Julius no-code data agents, Rows spreadsheet AI, and Perplexity cited market research.
• Best AI for Interview Prep is the June 26 verified adjacent guide for ethical prep, source-grounded resume/job-description research, spoken roleplay, company-source checking, and technical mock practice.
• Best AI Tools for Consultants is the June 27 verified adjacent guide for source-backed market, company, vendor, meeting, deck, and memo workflows.
• Best AI Tools for Journalists is the June 27 verified research-safety guide for cited web research, fixed source packs, source logs, academic claim checks, account security, and primary-source verification.
• Best AI Tools for Lawyers is the June 27 verified legal-research and drafting buyer guide for Harvey, Claude, Spellbook, CoCounsel Legal, and Lexis+ with Protege, with authority, citation, privilege, and matter-data checks foregrounded.
• Best AI Tools for Nonprofits is the June 27 verified nonprofit research and reporting guide for NotebookLM source packs, Google Workspace AI, ChatGPT discounts, Claude nonprofit pricing, grant workflows, and donor/beneficiary data safety.

AiPedia is intentionally not promoting the old medical-research guide until it is rebuilt with medical-specific sources, privacy guidance, regulator context, high-stakes safety caveats, and current life-sciences AI coverage.

Trust Rules for Research AI

Do not cite generated text. Cite the original source. Do not treat a linked source trail as proof of truth. Do not upload confidential research, private interviews, patient data, unpublished manuscripts, or restricted institutional material without checking policy and vendor terms. AI can shorten the path to evidence, but it cannot own the evidence judgment.