1. NeoMundi Observatory of Generative AI

A public program for the continuous measurement of generative systems.
Every two weeks, NeoMundi publishes a new observation cohort of generative AIs.

Each cohort systematically measures five dimensions: Stability, Validity, Information Density, Risk, and Cost Efficiency.
The goal is not only to compare models, but to make their behavior observable, measurable, and verifiable over time.

The results are accompanied by raw data, documented methodology, and verification links to promote auditability, comparison, and public debate.

2. Latest Publication

[Cohorte 02 – May 2026 – In preparation]

Archive of Cartographies

All Cartographies

3. Editorial Program

A public rhythm: one cohort every two weeks
Every two weeks, on Mondays, NeoMundi publishes a new observation cohort.
Each cohort systematically includes a global cartography and a synthetic reading of the main measured signals.

Depending on the wave, the publication may also include:

  • Intra-provider benchmarks,
  • segment-specific measurements,
  • singular case analyses,
  • methodological observations,
  • inter-cohort evolution comparisons.

4. The Five Measured Dimensions

Chaque cohorte mesure systématiquement cinq dimensions :

  • Stability: measurement of runtime stability during generation.
  • Validity: evaluation of alignment with verifiable references or benchmarks.
  • Information Density: measurement of the amount of useful information produced.
  • Risk: identification of signals of drift, instability, or problematic responses.
  • Cost Efficiency: Analysis of the relationship between performance, cost, latency, and operational efficiency.

5. Method, Data, and Auditability

An open, documented, and verifiable method

NeoMundi progressively documents and publishes:

  • The methodology
  • CSV / JSON datasets
  • Protocol versions
  • Source code (when available)
  • Interpretation limits

This progressive disclosure aims to make the observations comparable, debatable, and verifiable.

Read the methodology
See data on GitHub

6. Statement of Principle

Measuring does not mean certifying truth

NeoMundi does not claim that a runtime stability signal guarantees the accuracy of a response.

An AI can produce a correct answer while showing an unstable runtime signature.
It can also produce a false answer without any observable turbulence.

This is why the Observatory explicitly distinguishes between stability, validity, informational density, risk, and cost/performance efficiency.

Scroll to Top