Aircade retro glove controller header
← All models

SIQ-1-35B

SIQ-1-35B model image

SIQ-1-35B

credit: SIQ
B70 input tok/s33.1
B70 decode tok/s223.7
B70 per-agent tok/s7.8
B70 live fleet32 · Intel Arc Pro B70

Q5_K_M qwen35moe · llama.cpp SYCL on Intel Arc Pro B70 · context-calibrated team of 7 · 20-min timed production loop.

model dossier

Model sheet

Playable artifacts and benchmark context for this model page.

Road Hopper
2 builds
Robot-Filled Maze Shooter
2 builds
Ribbit Rush
0 builds
Builds
4 recorded outputs
Method
team 7 / 20-min build
Runtime
llama.cpp SYCL on Intel Arc Pro B70
Quant
Q5_K_M qwen35moe
Bench
B70 live: 32 agents / 223.7 decode tok/sec
benchmark sheet

B70 live throughput profile

B70 input tok/s33.1
B70 decode tok/s223.7
B70 per-agent tok/s7.8
B70 live fleet32 · Intel Arc Pro B70
aggregate decode tok/secper-agent tok/sechighlight = selected live fleet
tool · 1 agents
63.1 dec82.8/agent · 134.2 in
tool · 8 agents
141.9 dec19.7/agent · 95.2 in
tool · 16 agents
162.8 dec11.2/agent · 58.4 in
tool · 32 agents
223.7 dec7.8/agent · 33.1 in
structured · 32 agents
164.8 dec6.5/agent · 43.2 in
novel · 32 agents
125.3 dec5.8/agent · 55.0 in

Highlighted row is the selected 32-agent tool-prompt live fleet on Intel Arc Pro B70. Structured and novel rows are LocalMaxxing prompt classes.

Raw B70 live table
classagentsprefill_tpsagg_decode_tpsper_agent_tpsbatch_wall_sstatus
tool1134.263.182.84.06OK
tool895.2141.919.714.43OK
tool1658.4162.811.225.16OK
tool3233.1223.77.836.62OK
structured3243.2164.86.549.71OK
novel3255.0125.35.858.93OK
production sheet

Teamwork build

No fan-out: a 7-agent team collaborates on ONE game for a 20-minute timed loop — a lead assigns focus areas, workers improve in parallel, a merger integrates the best, validated each round so it never regresses. Team size comes from the context/teamwork Pareto (single-agent footprint × 1.10 → agents that fit the window). Methodology ↗.

GameScoreSeed → FinalAssertionsRoundsSeed → Final sizeBuild time
Road Hopper61.061.0 → 61.06/6127 → 27 KB20 min
Robot-Filled Maze Shooter71.071.0 → 71.07/7126 → 26 KB20 min

Generated game outputs

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

Road Hopper 2 versions

road_hopper_siq_b70_live · Road Hopper · team buildfullscreen ↗ · compare
road_hopper_siq_gb10_teamwork · Road Hopper · team buildfullscreen ↗ · compare
Road Hopper team build SIQ-1-35B road_hopper_siq_b70_livefullscreen ↗
Road Hopper team build SIQ-1-35B road_hopper_siq_gb10_teamworkfullscreen ↗

Robot-Filled Maze Shooter 2 versions

maze_siq_b70_live · Robot-Filled Maze Shooter · team buildfullscreen ↗ · compare
maze_siq_gb10_teamwork · Robot-Filled Maze Shooter · team buildfullscreen ↗ · compare
Robot-Filled Maze Shooter team build SIQ-1-35B maze_siq_b70_livefullscreen ↗
Robot-Filled Maze Shooter team build SIQ-1-35B maze_siq_gb10_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.