InclusionAI’s ZwZ2-9B is a vision-and-video model shaped for deployment; Liquid AI’s LFM2.5 2.6B Base is a text checkpoint intended for building on. OrcaRouter’s comparison makes the useful point that model choice starts with the job, not the parameter count.
Watch Desk analysis
What happened
OrcaRouter says InclusionAI posted ZwZ2-9B on 16 October. Its account describes a 9.41-billion-parameter checkpoint with image and video support, while LFM2.5 2.6B Base is a 2.697-billion-parameter, text-only checkpoint that Liquid AI presents as a starting point for fine-tuning.
There are practical differences beyond size. OrcaRouter says ZwZ2-9B has no model card, evaluation results, training information or declared licence in the repository it examined. It says LFM2.5 Base has a model card and uses Liquid AI’s LFM Open License v1.0. OrcaRouter’s comparison also describes different architectures and context ceilings, but those numbers do not turn the models into like-for-like alternatives.
Why it matters
A ready-to-deploy multimodal checkpoint and a base model for custom training answer different needs. Treating them as a head-to-head size contest could send a team towards a model that cannot handle its inputs, or one that still needs substantial development before it behaves like an assistant.
The documentation gap matters too. OrcaRouter’s account says key details about ZwZ2-9B, including its licence and evaluation results, are absent from the repository. That leaves prospective users with basic homework before they can make a confident deployment decision.
Our read
This is a useful comparison precisely because it refuses to crown a winner. If you need image or video input, ZwZ2-9B may be the more relevant shape of model; if you want a small text checkpoint to adapt, LFM2.5 Base is the more obvious starting point. But the missing documentation around ZwZ2-9B is not a minor footnote. Don’t mistake a plausible technical profile for a complete deployment brief.
What to watch
- Whether InclusionAI publishes a model card, licence and evaluation results for ZwZ2-9B.
- How the checkpoint performs on independently described vision and video tasks.
- Whether teams use LFM2.5 Base as intended, with the additional tuning and evaluation that entails.
Discussion spark: Would you deploy a promising model before its licence and evaluation results are documented, or is that a line teams should not cross?
Sources and evidence
- ZwZ2-9B vs LFM2.5 2.6B Base: These Never Competed at All – OrcaRouter (16 October 2026, 12:55 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.