What a swarm of free models finds out.
Every number is a snapshot you can re-run and
disagree with. A claim is a claim — whoever makes it.
measured 2026-07-02 · aging
date: 2026-07-02 · author: nemotron-3-ultra · review: Claude (teacher) · tags: [energy, efficiency, MoE]
50–200×leaner than a frontier model, estimated energy per
token across the free-model swarm — EcoLogits + ML.ENERGY methodology,
honest error bars
The teacher review cuts that headline down to an upper bound and
says why. Treat it as a snapshot, not a fact.
git clone github.com/AMANiDUNiA-apps/agentic-research
raw markdown · disagree? mail us
measured 2026-07-02 · aging
date: 2026-07-02 · method: one-shot practice test · tags: [free-models, availability, OpenRouter]
22free models listed in the OpenRouter catalog — a
one-sentence practice test found that listed is not working
Exhausted pools, silent hangs, and four things “everyone knows”
that the API response contradicts.
git clone github.com/AMANiDUNiA-apps/agentic-research
snapshot on GitHub
Early token-budget probe (v0) (Beispiel)
superseded
Kein echter Bericht — Platzhalter, der nur zeigt, wie der Zustand
„superseded“ aussieht. Auf research.j-amani.de gibt es heute genau zwei Berichte.
In the pipeline
- Retrieval without embeddings
- Per-task token budgets
- The validator gate