Models
This page lists every VLA model Argos supports or is evaluating, with what's needed to run each one and how it performs on real hardware.
smolvla
Supported450M params
vla.cpp's "golden reference" architecture — every other arch is validated against this recipe.
pi0
Candidate3B params
Needs `huggingface-cli login` + accepting the PaliGemma license before first use.
bitvla
Candidate2.4B params
1-bit/ternary weights, smallest footprint of any served model (1.3 GiB resident on Orin Nano) despite being 2.4B params.
evo1
Candidate770M params
Second-highest success rate of the 7 (94.5% on LIBERO-Object, RTX 3060), fits comfortably on Orin Nano.
gr00t-n1.5
Candidate3B params
V1.x candidate — filename now confirmed, launch args verified
gr00t-n1.6
Not compatible3B params
~6GB resident weights exhaust the 8GB unified memory pool entirely — does not fit even with the simulator offloaded elsewhere.
gr00t-n1.7
Not compatible3B params
Highest success rate of all 7 archs on LIBERO-Object (98.0%, RTX 3060), but same ~6GB footprint problem as N1.6 — does not fit the 8GB target hardware tier at all.