Claude Science vs ThesisAI - Why ThesisAI is Better for Writing

What is Claude Science? Anthropic's Research Workbench Explained

Claude Science is a downloadable application for macOS, Windows and Linux. It carries no separate price, but it requires a paid Claude plan - Pro, Max, Team or Enterprise. Principal investigators at academic and nonprofit research institutions can apply for a free Claude Team plan for scientists that covers their lab for a year.

Work is organised into Projects (a research topic) and Sessions (an individual line of analysis). Each project keeps its own memory, data sources, installed skills and artifacts. Inside a session you type a plain-language request, and a coordinating agent decides what needs doing and spins up specialist sub-agents to do the parts.

The core capabilities are:

• Reproducible artifacts: every figure, table and result ships with the exact code, the environment that produced it, and the full message history behind it
• A reviewer agent that flags incorrect citations, numbers it cannot trace, and figures that do not match their underlying code
• More than 60 scientific database connectors, including UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, GEO and OpenAlex
• Compute that runs locally, over SSH to a remote Linux box or a Slurm HPC cluster, or on Modal for GPU work
• Persistent Python and R kernels that keep variables in memory across sessions
• Native renderers for 3D protein structures, genome browser tracks, chemical structures, sequence alignments and tabular data
• Pre-configured modules for genomics, single-cell RNA-seq, proteomics, structural biology, cheminformatics, phylogenetics and CRISPR screens
• Extensibility through MCP connectors and reusable skills saved as SKILL.md files

Data stays on your own infrastructure. Anthropic's stated position is that your datasets never leave your machine and only the context window is transmitted, which matters for anyone working under a data governance agreement.

Read our full write-up in the blog post Claude Science Review: What Anthropic's Research Workbench Actually Does.

What Claude Science Does Well

Its strongest idea is not an AI idea at all - it is a reproducibility idea. In a normal AI-assisted analysis you get a figure and a wall of code in a chat transcript, and reconstructing what produced what is your problem six months later. Claude Science inverts this: the figure is a bundle containing its code, its environment and the conversation that produced it, so you can reopen it much later and see exactly what was run.

The second strength is the reviewer agent. It inspects outputs as they appear and compares what the main agent said happened against what the execution record shows did happen. When a figure caption claims a filter was applied and the code shows it was not, that mismatch is precisely what it catches.

Third, compute handling is careful. The app drafts a plan before reaching for a new resource and waits for your approval before running anything compute-intensive or billable, rather than quietly burning credits in a loop.

Researchers using it in the field back this up. A neuroscientist at UCSF found it suggested analysis steps he had not considered on glioma genetics datasets. A structural biologist at EPFL singled out the transparency, describing an app that shows you step by step what it is doing, which code it is using, and why. A researcher at the Allen Institute built a multi-agent template that assembled literature reviews running past 100 pages with automated citation verification.

For a computational researcher with data and a cluster, this is a serious piece of engineering and worth installing.

Why Claude Science Falls Short for Thesis Writing

Four limits decide whether Claude Science belongs in a thesis workflow, and none of them is a bug. They are the consequences of a product that chose an audience and built properly for it.

It does not generate the document

Claude Science drafts methods, results and figure captions alongside the analysis they describe, with Markdown and LaTeX preview. That is genuinely useful, and it is also the narrow slice of a thesis that sits closest to the code. There is no workflow that takes you from a research question to a structured 80-page document. It does not plan chapters, build an argument across sections, or fill a blank page.

The domain coverage is life sciences

Genomics, proteomics, structural biology, cheminformatics. If you are in economics, psychology, education, business, law or history, the pre-configured infrastructure that makes this app impressive simply is not there. What you get is a capable general model attached to a code runner - useful, but not a research workbench for your field.

There is no reference library

You can upload documents into a session, but there is no persistent library across projects, no Zotero or Mendeley import, and no citation style engine. The citation checking it performs verifies that a cited claim traces to something real in the run. That is not bibliography management, and it will not produce a formatted reference list in your department's required style.

The reviewer does not check your science

This boundary is worth understanding properly, because it is the difference between useful and dangerous. The reviewer does not re-run your analysis and it does not evaluate whether the analysis was the right one. It verifies that the narrative matches the execution. A methodologically wrong analysis, faithfully described, passes review.

Researchers have raised the same point in stronger terms: scientific methodology is not in the training objective. There is also no PRISMA screening, no dual-reviewer blinding, and no structured extraction with quote-level provenance, so the app does not produce the audit trail a systematic review chapter needs. It cannot read paywalled papers either, which is the same wall every literature tool hits.

Claude Science vs ThesisAI: The Real Difference

These two tools sit at opposite ends of the same project, which is why the comparison is less a competition than a division of labour.

Claude Science starts from your data and your compute, and its output is a set of reproducible artifacts: code, figures, tables, and the record of how they were made. It assumes you already know what you want to investigate and how the finished document should look.

ThesisAI starts from your topic and your sources, and its output is the document. From a single prompt it drafts a full academic text of up to 80 pages with inline citations, chapter structure and logical flow, then exports to LaTeX, Word, PDF and BibTeX. You can upload up to 500 papers, import your library from Zotero or Mendeley, and push straight into Overleaf.

The citation models are different in a way that matters. Claude Science verifies that cited claims trace back to the execution record. ThesisAI retrieves papers from indexed academic databases and verifies each inline citation against the paper it came from, so the reference list contains work your examiner can actually open.

In practical terms:

• ThesisAI is better for students and researchers whose deliverable is a written thesis, dissertation or paper, in any discipline.
• Claude Science is better for computational and wet-lab researchers who need to run, document and reproduce analyses on their own data.

If you are a computational PhD student, the honest answer is to use both: Claude Science for the analysis chapters, where reproducible artifacts are worth real effort, and ThesisAI for the document that has to be written around them.

Final Verdict: Who Should Use Claude Science?

Claude Science is a good fit if:

• You write code to analyse data as a routine part of your research
• You work in genomics, proteomics, structural biology or cheminformatics
• You have access to a cluster or GPU compute
• You have ever failed to reproduce your own result from six months ago
• Your PI can put the lab on the free academic Team plan

ThesisAI is the better choice if:

• Your bottleneck is writing and structuring, not computing
• You are starting from a blank page and need chapters, flow and an argument
• Your research is qualitative, theoretical, or outside the life sciences
• You need a verified reference list in a specific citation style
• You want something that works in a browser without installing an environment

Claude Science is the first AI research product that looks like it was designed by people who have actually fought with a conda environment at 2am. The reproducibility model is the right one and the reviewer is a sensible, appropriately modest idea. It is also aimed squarely at people with data and a cluster - and a thesis is mostly not that.