Aircade retro glove controller header
← All models

DeepSeek V4 Pro

DeepSeek V4 Pro model image

DeepSeek V4 Pro

credit: DeepSeek

thinking mode · OpenCode single agent · 1 xhigh subagents split across the two game prompts.

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
1 xhigh subagents
Runtime
OpenCode single agent
Quant
thinking mode
Bench
No local throughput sweep recorded

Generated game outputs

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

Road Hopper 1 versions

froggai_deepseek_v4_pro · Road Hopper · FroggAI deepseek evalfullscreen ↗ · compare
Road Hopper FroggAI deepseek eval DeepSeek V4 Pro froggai_deepseek_v4_profullscreen ↗

Robot-Filled Maze Shooter 1 versions

maze_deepseek_v4_pro · Robot-Filled Maze Shooter · Maze deepseek evalfullscreen ↗ · compare
Robot-Filled Maze Shooter Maze deepseek eval DeepSeek V4 Pro maze_deepseek_v4_profullscreen ↗
UNDER RAPID CONSTRUCTION - WILL CHANGE
Frostmark retro console footer
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.