Compare

Odins vs LightweightMMM

LightweightMMM was Google's open-source Bayesian MMM library. Google archived the repository on 19 January 2026, and its README says the library is no longer supported and recommends Meridian. If your model runs on it, it needs a new home. Here is how Odins compares.

Book a demo See the comparison

Reviewed October 2026 · sources cited on this page

Odins

Managed marketing investment platform

Bayesianfully manageddaily datamonthly retrain

LightweightMMM

Archived open-source library

BayesianNumPyro and JAXarchived January 2026

At a glance

Side by side.

The short version of the whole page. Details and sources below.

Odins

LightweightMMM

What it is

OdinsA managed marketing investment platform: data, model, recommendations and a team behind them

LightweightMMMAn open-source Python library for Bayesian marketing mix models, released by Google and not an official Google product

Status

OdinsAn operated platform, maintained by our team

LightweightMMMArchived and read-only since 19 January 2026; the README says it is no longer supported; last release 0.1.9 in May 2023

Methodology

OdinsBayesian MMM, one model per market, priors set with your team in 2 to 3 working sessions

LightweightMMMBayesian MMM built on NumPyro and JAX, national or geo-level, with adstock, hill-adstock or carryover transforms

Who runs it

OdinsOur team: data integration, model build and maintenance, and review of every recommendation

LightweightMMMYour own data scientists, end to end

Cadence

OdinsDaily data flow, monthly retrain and reviewed recommendation; scenarios and forecasts any time

LightweightMMMWhatever your team builds and maintains around it

Offline and drivers

OdinsTV, radio, OOH and print plus pricing and promotions in the same model

LightweightMMMAny media you supply as data, plus extra features such as price, which must be known in advance for optimization

Budget optimization

OdinsBudget size and allocation with confidence ranges, scenarios and forecasts any time in the platform or from Claude via the Odins MCP

LightweightMMMAn optimizer that reallocates a fixed total budget across channels

Cost

OdinsSubscription priced for the budget it optimizes

LightweightMMMFree under the Apache 2.0 license, paid for in staff time

Time to value

OdinsFirst verified model in 4 to 6 weeks from kickoff, data collection and structuring included

LightweightMMMDepends on your team; no new fixes or releases will come

LightweightMMM facts: github.com/google/lightweight_mmm (README and archive notice) and pypi.org/project/lightweight-mmm, read October 2026.

The honest read

Credit first: LightweightMMM put Bayesian MMM in the open.

It gave marketing teams a free Bayesian marketing mix model they could read line by line and run on their own data. Meridian builds on the same idea.

A Bayesian MMM free under the Apache 2.0 license
National and geo-level models, the geo version hierarchical across regions
Three media transforms, adstock, hill-adstock and carryover, to fit how a channel decays
A built-in optimizer for splitting a fixed budget across channels

The difference

Moving off LightweightMMM: rebuild it, or hand it over.

There are two routes. Rebuild on Google Meridian, its successor, with your own data scientists. Or hand the model to a team that runs it for you. Odins is the second route: one Bayesian model per market, built on the same statistical principles, run end to end.

No code to port: we connect the data, set priors with your team in 2 to 3 working sessions, and build one model per market
TV, radio, OOH and print in the same model as digital, next to pricing and promotions
Data flows in daily and scenarios and forecasts run any time; the model is retrained monthly and a person reviews each recommendation
Forecasts stored before actuals arrive and checked at each retrain, so the model's record is there to inspect

Channel contribution, 80% ranges

from the posterior

Search 14 to 16% optimize

Social 9 to 13% optimize

Display 2 to 9% test first

A tight range is a decision. A wide range is a test plan, with cost boundaries.

The choice

Who should choose which.

Your LightweightMMM model will keep running for now. The question is who maintains what comes next.

Choose LightweightMMM if

Your data science team can keep a frozen library running and owns the model fully
The model is a research tool, not the basis for budget decisions
You plan to move to Meridian and want the old model as a reference until then

Choose Odins if

You want the budget decision answered without rebuilding the model in-house
TV, radio, OOH or print carry real budget, or sales are one national weekly series where priors carry the model
You want a person to review every recommendation, and scenarios you run yourself any time

The honest overlap

We build on the same Bayesian principles LightweightMMM made accessible. The difference is everything around the model: data, maintenance, review and verification. Ask any vendor, or your own team, who will maintain the model in two years.

FAQ

Frequently asked questions.

Is LightweightMMM still supported?

No. Google archived the repository on 19 January 2026, and the README says LightweightMMM is no longer supported and recommends Meridian. The last release on PyPI is 0.1.9, from May 2023. The code is still available, but no fixes or updates will come.

What replaced LightweightMMM?

Google Meridian, released in January 2025 as Google's official Bayesian MMM. If you want to rebuild in-house, it is the natural next step. Our page on Google Meridian: build vs buy covers what that takes, and Odins vs Google Meridian compares the two routes.

Can we migrate our LightweightMMM model to Odins?

We do not port the code. We start from your data: spend, sales, offline media and business drivers. What your team learned from the old model goes into the priors, which we set with you in 2 to 3 working sessions. A first verified model is ready 4 to 6 weeks from kickoff, data collection and structuring included.

How much does it cost to replace LightweightMMM?

Meridian is free to license and paid for in staff time, like LightweightMMM was. Odins is a subscription priced for the budget it optimizes, with the team included and no data science hires needed.

Is Odins built on LightweightMMM or Meridian?

No. Odins runs its own Bayesian model, built on the same statistical principles as the open frameworks and operated end to end as a product.

What are the best alternatives to LightweightMMM?

It depends on what you need from the model. We compared five options, with sources and who each one fits, on a separate page: the best LightweightMMM alternatives. Odins is on the list, and we say so up front.