Research and data

AI model catalog

Explore model specifications and reported benchmark results in our model comparisons. Scores come from the sources identified alongside the data. They are not independent experiments conducted by TheQuery.

Download model data (JSON) · Download benchmark observations (CSV)

Versioned snapshots and citation metadata on GitHub

The downloads preserve evaluator, source, benchmark version, harness, tools, reasoning effort, and evaluation date where recorded. Missing values mean unknown, not zero or disabled.

Methodology and limitations

The catalog compiles published model specifications and evaluation claims. Multiple observations can exist for one model and benchmark. A shared benchmark name does not establish comparable conditions: check the version, harness, tool access, evaluator, and reasoning budget before comparing scores.

Export timestamps identify when a snapshot was retrieved, not when a result was independently reproduced. Source links provide provenance, not a guarantee that the source remains available or that a vendor claim is correct. Coverage is selective and does not represent every model or evaluation.

Read How to Audit AI Benchmark Claims: A Practical Guide for the checks to apply before drawing conclusions.

Free technical books

• RAG + Knowledge Graph Master Course

  From beginner to hire-ready enterprise AI engineer. Master RAG, Knowledge Graphs, and hybrid retrieval systems for production-grade AI applications.

• AI: From First Principles

  A comprehensive guide for developers who want to actually understand AI. Covers everything from what AI really is, through data foundations, math essentials, classical ML, neural networks, transformers, modern AI systems, and production deployment.

Both books can be read without an account. Each landing page includes a suggested citation and BibTeX.

Citing and correcting the data

Cite TheQuery, “AI model catalog,” https://www.thequery.in/research, and your retrieval date. Also cite the original evaluator and source for any individual score. Keep the downloaded snapshot so readers can reproduce which data you used.

Source materials retain their original rights. No new license for third-party material is implied. For corrections, send the model, benchmark, source URL, and proposed correction to [email protected].