← Guides

AI for Science (2026): The Best AI Tools for Scientists, Compared, and an Agent That Runs the Analysis End to End

How scientists use AI in November 2026: discovery models like AlphaFold, AI scientist agents like Edison's Kosmos, literature tools like Elicit and Consensus, general assistants, prices compared, and the lab workflows SPIRITT runs from data to paper.

Flat risograph illustration of a smiling scientist in a lab coat at a microscope with flasks and test tubes while a friendly robot holds a laptop with a rising chart and a DNA helix floats above, navy and orange on cream

AI for science means using AI to read the literature, analyze data, predict structures and properties, generate hypotheses and automate lab paperwork. In November 2026 scientists combine four kinds of tools: discovery models such as AlphaFold 3 (free on AlphaFold Server for non-commercial research), AI scientist agents such as Edison Scientific's Kosmos, literature tools such as Elicit and Consensus, and general assistants such as ChatGPT, Claude and Gemini. SPIRITT is an AI agent that runs the scientist's whole workflow end to end: it searches and reads papers, codes and runs your analysis in its own cloud computer, makes figures, drafts the paper and grant sections, and handles lab admin, with every step logged. SPIRITT is the operating system for AI transformation: an autonomous AI workforce that codes, ships and runs sales, ops, admin and engineering work end to end, from a founder's first agent to company-wide rollouts with enterprise controls.

This guide is for working scientists, lab managers and R&D teams. For a broader look at research tools across all fields, see AI for research.

How AI is changing science in 2026

Three milestones explain why "AI for scientists" became a mainstream search:

  • Structure prediction became routine. The 2024 Nobel Prize in Chemistry went to Demis Hassabis and John Jumper for AlphaFold's protein structure prediction and to David Baker for computational protein design. AlphaFold 3 extends prediction to complexes with DNA, RNA and small molecules.
  • AI scientist agents arrived. Edison Scientific, a commercial spinout of FutureHouse, launched Kosmos in November 2025: an agent that takes a research goal and datasets, reads literature, writes and runs analysis code and returns a cited report.
  • Governments made AI for science a priority. In November 2025 the US Department of Energy launched the Genesis Mission to build AI platforms for science across the national laboratories.

For a typical lab the practical change is smaller and more useful: the hours spent searching papers, cleaning data, writing analysis scripts and formatting figures can now be handed to AI and checked by a person.

The best AI tools for scientists, compared (November 2026)

ToolKindBest forPrice (US)
AlphaFold ServerDiscovery modelPredicting protein structures and complexesFree for non-commercial research
Edison Scientific (Kosmos and agents)AI scientist agentsLong autonomous investigations; literature, analysis and precedent agentsPlayground: 10 free credits a month; new Kosmos access by application, prices not published
ElicitLiterature toolSystematic reviews and data extraction from papersFree basic plan; Pro $49 per user a month billed yearly
ConsensusLiterature toolQuick answers from peer-reviewed studiesFree; Pro $12 a month billed yearly ($20 monthly)
Gemini NotebookSource-grounded notesAnswers and audio overviews from your own papersFree; more with Google AI Pro $19.99 a month
Claude and ChatGPTGeneral assistantsExplaining methods, writing, code for analysisClaude Pro $20 ($17 billed yearly); ChatGPT Plus $20
SPIRITTAI agentThe whole workflow: literature, analysis code, figures, drafts, lab admin, monitoringFree to start; Starter $32, Pro $256, Max $1,024 a month

Prices are vendor list prices checked in November 2026. Edison Scientific retired its founding Kosmos subscription and is rolling out new access, so check its site for current terms.

Discovery models

Discovery models predict things a lab would otherwise measure: protein structures, binding, material properties, weather. They are specialist tools, used through a web server, an API or open weights, and their predictions are hypotheses to test, not results.

AI scientist agents

AI scientist agents run a long investigation on their own: read hundreds of papers, analyze a dataset, propose and test hypotheses in code and write a report. They suit high-value questions where a run of several hours is worth it, and they need expert review of every conclusion.

Literature tools

Elicit and Consensus search databases of academic papers rather than the open web, which makes their answers easier to trace to peer-reviewed studies. Elicit's extraction tables are a strong fit for systematic reviews.

Seven scientific workflows SPIRITT runs end to end

SPIRITT is an AI agent with its own isolated cloud computer, a browser, a terminal and 1,500+ app connections. For a lab, that means it can do the work of a research assistant, a data analyst and a lab coordinator, and keep the code and sources for every result.

  1. Literature review to extraction table. Defines the search, reads full texts you can access, extracts methods, sample sizes and effect sizes into a sheet with sources, and drafts the review.
  2. Data analysis with real code. Cleans instrument and survey data, runs the right statistical tests in Python or R, and makes publication-quality figures. You get the notebook, so every number can be rerun.
  3. Pipelines for repeat experiments. Turns a one-off script into a pipeline that processes each new plate, run or batch the same way and posts a summary to Slack.
  4. Papers and grants. Drafts methods and results from your analysis, formats references, checks figure legends, and helps with proposals. See AI for grant writing.
  5. Lab operations. Tracks reagents and orders in a sheet, drafts purchase requests for approval, schedules equipment and summarizes lab meetings.
  6. Tools the lab is missing. Codes and hosts a small app: a sample tracker, a protocol library, a dashboard for an ongoing study.
  7. A living watch on your field. Checks for new papers, preprints and datasets on your topic every week and sends a short digest with why each one matters.

Long jobs keep running on SPIRITT's cloud computer after you close the laptop; see AI agents for long-running tasks. SPIRITT composes models from Anthropic, OpenAI, Google, xAI and others inside each task, plus its own Kay models, so reading, reasoning, coding and writing each get the strongest mix. It can also hand a structure question to a specialist tool and bring the result back into your analysis.

A worked example: a plate-reader study

A cell biology lab runs a 96-well viability assay every week. SPIRITT reads the plate-reader export, flags two wells with pipetting outliers, normalizes to controls, fits dose-response curves in Python, reports IC50 values with confidence intervals, and produces the same figure layout every week. A postdoc reviews the summary on Slack and approves the figures for the lab meeting. The example is a composite with made-up details.

Keeping AI-assisted science rigorous

  • Keep the code. Every analysis should be reproducible from data and script, not from a chat transcript.
  • Verify citations in the source. AI tools summarize; the paper is the evidence.
  • Pre-register what you can. Decide tests and endpoints before asking AI to analyze, to avoid fishing.
  • Protect sensitive data. Use tools that keep data isolated and do not train on it, and follow your institution's rules for human-subject data.
  • Disclose AI use. Many journals and funders now ask how AI was used; keep a log. SPIRITT logs every action in an audit trail.

FAQ

What is the best AI for science?

It depends on the task. AlphaFold Server is the standard for protein structures, Elicit and Consensus for literature, Edison Scientific's agents for long autonomous investigations, and Claude or ChatGPT for explaining methods and writing. SPIRITT is best when you want one agent to run the workflow from papers to analysis to draft.

What is the best AI for scientists?

Most scientists combine a literature tool, a general assistant and a way to run analysis code. SPIRITT brings those together: it reads papers, writes and runs your analysis, makes figures and drafts the write-up, with the code kept for review.

Can AI make scientific discoveries?

AI has already changed fields such as structural biology, and AI scientist agents now propose and test hypotheses in code. Discoveries still need experiments and expert judgment; treat AI output as a hypothesis or a draft until it is validated.

What is an AI scientist?

An AI scientist is an agent that runs a research loop on its own: reading literature, analyzing data, generating hypotheses and writing up results. Edison Scientific's Kosmos is a well-known example.

Is there free AI for scientific research?

Yes. AlphaFold Server is free for non-commercial research, Elicit and Consensus have free plans, Edison's Playground gives 10 free credits a month, and ChatGPT, Claude and Gemini have free tiers. SPIRITT is free to start with daily credits.

Can AI analyze my experimental data?

Yes. SPIRITT reads your exports, checks for problems, runs the statistics in Python or R, makes figures and explains each step, and keeps the notebook so you can rerun or change the analysis.

How do scientists use AI in the lab?

For literature reviews, data cleaning and analysis, figure preparation, writing papers and grants, protocol and inventory management, and keeping up with new publications.

How we checked

SPIRITT publishes this guide. Edison Scientific details come from edisonscientific.com, its Kosmos announcement of November 5, 2025 and independent coverage of its 2026 access changes. AlphaFold details come from Google DeepMind's AlphaFold Server pages; the Nobel Prize details from the Nobel Foundation's 2024 announcement; the Genesis Mission from the US Department of Energy. Elicit and Consensus prices come from our November 2026 AI for research comparison; Claude prices from claude.com/pricing; ChatGPT prices from OpenAI's pricing documentation; Google AI prices from Google One, all read in November 2026. SPIRITT capabilities and prices come from spiritt.ai and the live SPIRITT plan catalog. We read the top Google results for "ai for scientists" on November 22, 2026. The plate-reader example is a composite with made-up details, not a real customer.

Hand your analysis to an agent

Buy from builders who use what they sellBuilt usingSPIRITT