Aircade retro glove controller header
← All models

Frosty Ornith-1.0-35B

Frosty Ornith-35B model image

Frosty Ornith-1.0-35B

credit: Frosty40 / Ornith
input tok/s2,352.6
decode tok/s277.1
per-agent tok/s8.7
agent Pareto32 · NVIDIA GB10

GGUF-NVFP4 (qwen35moe) · llama.cpp · selected fleet 32 agents (its Pareto knee)  ·  Download on Hugging Face ↗. Full throughput curve on Benchmarks.

model dossier

Model sheet

Playable artifacts and benchmark context for this model page.

Road Hopper
7 builds
Robot-Filled Maze Shooter
10 builds
Ribbit Rush
0 builds
Builds
17 recorded outputs
Method
Pareto fleet 32
Runtime
llama.cpp
Quant
GGUF-NVFP4 (qwen35moe)
Bench
32 agents / 277.1 decode tok/sec
benchmark sheet

Throughput profile

input tok/s2,352.6
decode tok/s277.1
per-agent tok/s8.7
agent Pareto32 · NVIDIA GB10
aggregate decode tok/secper-agent tok/sechighlight = selected fleet
1 agents
69.0 dec69.0/agent · 2,286.4 in
8 agents
206.5 dec25.8/agent · 2,372.4 in
16 agents
225.7 dec14.1/agent · 2,356.0 in
32 agents
277.1 dec8.7/agent · 2,352.6 in
64 agents
326.3 dec5.1/agent · 2,351.4 in
96 agents
342.6 dec3.6/agent · 2,355.6 in
128 agents
350.6 dec2.7/agent · 2,353.2 in
192 agents
361.5 dec1.9/agent · 2,358.0 in
256 agents
372.3 dec1.4/agent · 2,356.2 in

Highlighted row is the selected build fleet / Pareto knee.

Raw sweep table
agentsprompt_proc_t/sagg_gen_t/sper_agent_gen_t/stotal_t/sreply_latency_256tok_s
12286.4068.9868.98195.153.711
82372.40206.4825.81527.609.919
162355.98225.7414.11568.3118.145
322352.62277.158.66672.9129.558
642351.41326.315.10766.2550.210
962355.60342.593.57796.1871.736
1282353.23350.622.74810.3793.458
1922357.97361.531.88830.05135.957
2562356.25372.281.45848.67176.038

Generated game outputs

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

Road Hopper 7 versions

road_hopper_ornith_teamwork · Road Hopper · teamwork evalfullscreen ↗ · compare
road-hopper_ornith_00 · Road Hopper · v1fullscreen ↗ · compare
road-hopper_ornith_03 · Road Hopper · v4fullscreen ↗ · compare
road_hopper_ornith_b70_live · Road Hopper · B70 live teamwork evalfullscreen ↗ · compare
road_hopper_ornith_b70_stock · Road Hopper · B70 stock baselinefullscreen ↗ · compare
road-hopper_ornith_01 · Road Hopper · v2 · brokenfullscreen ↗ · compare
road-hopper_ornith_02 · Road Hopper · v3 · brokenfullscreen ↗ · compare
Road Hopper teamwork eval Frosty Ornith-35B road_hopper_ornith_teamworkfullscreen ↗
Road Hopper v1 Frosty Ornith-35B road-hopper_ornith_00fullscreen ↗
Road Hopper v4 Frosty Ornith-35B road-hopper_ornith_03fullscreen ↗
Road Hopper B70 live teamwork eval Frosty Ornith-35B road_hopper_ornith_b70_livefullscreen ↗
Road Hopper B70 stock baseline Frosty Ornith-35B road_hopper_ornith_b70_stockfullscreen ↗
Road Hopper v2 · broken Frosty Ornith-35B road-hopper_ornith_01fullscreen ↗
Road Hopper v3 · broken Frosty Ornith-35B road-hopper_ornith_02fullscreen ↗

Robot-Filled Maze Shooter 10 versions

maze_ornith_00 · Robot-Filled Maze Shooter · v1fullscreen ↗ · compare
maze_ornith_01 · Robot-Filled Maze Shooter · v2fullscreen ↗ · compare
maze_ornith_02 · Robot-Filled Maze Shooter · v3fullscreen ↗ · compare
maze_ornith_03 · Robot-Filled Maze Shooter · v4fullscreen ↗ · compare
maze_ornith_04 · Robot-Filled Maze Shooter · v5fullscreen ↗ · compare
maze_ornith_05 · Robot-Filled Maze Shooter · v6fullscreen ↗ · compare
maze_ornith_b70_failed_seed · Robot-Filled Maze Shooter · B70 failed attemptfullscreen ↗ · compare
maze_ornith_b70_live · Robot-Filled Maze Shooter · B70 live teamwork evalfullscreen ↗ · compare
maze_ornith_b70_stock · Robot-Filled Maze Shooter · B70 stock baselinefullscreen ↗ · compare
maze_ornith_teamwork · Robot-Filled Maze Shooter · teamwork evalfullscreen ↗ · compare
Robot-Filled Maze Shooter v1 Frosty Ornith-35B maze_ornith_00fullscreen ↗
Robot-Filled Maze Shooter v2 Frosty Ornith-35B maze_ornith_01fullscreen ↗
Robot-Filled Maze Shooter v3 Frosty Ornith-35B maze_ornith_02fullscreen ↗
Robot-Filled Maze Shooter v4 Frosty Ornith-35B maze_ornith_03fullscreen ↗
Robot-Filled Maze Shooter v5 Frosty Ornith-35B maze_ornith_04fullscreen ↗
Robot-Filled Maze Shooter v6 Frosty Ornith-35B maze_ornith_05fullscreen ↗
Robot-Filled Maze Shooter B70 failed attempt Frosty Ornith-35B maze_ornith_b70_failed_seedfullscreen ↗
Robot-Filled Maze Shooter B70 live teamwork eval Frosty Ornith-35B maze_ornith_b70_livefullscreen ↗
Robot-Filled Maze Shooter B70 stock baseline Frosty Ornith-35B maze_ornith_b70_stockfullscreen ↗
Robot-Filled Maze Shooter teamwork eval Frosty Ornith-35B maze_ornith_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.