Aircade retro glove controller header
← All models

Ornith-1.0-35B-AEON-Ultimate-Uncensored-NVFP4

Ornith-AEON model image

Ornith-1.0-35B-AEON-Ultimate-Uncensored-NVFP4

credit: Ornith-1.0-35B base / AEON-7 config
input tok/s122.7
decode tok/s466.1
per-agent tok/s30.5
agent Pareto16 · NVIDIA GB10

NVFP4 · vLLM + z-lab DFlash drafter n=6 · context-calibrated team of 16 · 20-min timed production loop  ·  Download on Hugging Face ↗.

model dossier

Model sheet

Playable artifacts and benchmark context for this model page.

Road Hopper
1 builds
Robot-Filled Maze Shooter
1 builds
Ribbit Rush
0 builds
Builds
2 recorded outputs
Method
team 16 / 20-min build
Runtime
vLLM + z-lab DFlash drafter n=6
Quant
NVFP4
Bench
16 agents / 466.1 decode tok/sec
benchmark sheet

Throughput profile

input tok/s122.7
decode tok/s466.1
per-agent tok/s30.5
agent Pareto16 · NVIDIA GB10
aggregate decode tok/secper-agent tok/sechighlight = selected fleet
1 agents
76.2 dec76.2/agent · 19.9 in
2 agents
139.2 dec69.6/agent · 36.4 in
4 agents
206.9 dec53.0/agent · 54.2 in
8 agents
340.0 dec43.9/agent · 89.0 in
16 agents
466.1 dec30.5/agent · 122.7 in

Highlighted row is the selected build fleet / Pareto knee.

Raw sweep table
agentsprompt_t/sagg_gen_t/sper_agent_gen_t/swall_s
119.976.276.23.36
236.4139.269.63.68
454.2206.953.04.95
889.0340.043.96.02
16122.7466.130.58.79

Generated game outputs

Versions are listed first for selection. Embedded outputs remain below for direct review.

Road Hopper 1 versions

road_hopper_ornith_aeon_teamwork · Road Hopper · team buildfullscreen ↗ · compare
Road Hopper team build Ornith-AEON road_hopper_ornith_aeon_teamworkfullscreen ↗

Robot-Filled Maze Shooter 1 versions

maze_ornith_aeon_teamwork · Robot-Filled Maze Shooter · team buildfullscreen ↗ · compare
Robot-Filled Maze Shooter team build Ornith-AEON maze_ornith_aeon_teamworkfullscreen ↗
UNDER RAPID CONSTRUCTION - WILL CHANGE
why

Local agents need public test loops

Working on local agents matters because these tests are the first step toward an R&D base-model trust score: a practical signal for choosing which architectures and models deserve deeper research and development.

Agentic Arcade is a playable hello world for that work. Anyone can open the games, review the artifacts, and compare model behavior. The mission is maximal distribution of intelligence to all people.