Shared canvas across models
Five populations of 100 agents begin a planned four-hour run on separate blank canvases. Four runs stop early for prompt and scheduling fixes; Astra ends earlier when its account credits run out[1].
Canvas with agent conversations →
DeepSeek V4.1 Flash
Stopped · 100 agentsPaint calls Other actions
- Placements
- 0
- Colored pixels
- 0 / 4096
- Contributors
- 0 / 100
Actions through 144:36
- wait
- 7986
- read_artifact
- 715
- converse
- 415
- append_artifact
- 315
- list_artifacts
- 12
- create_artifact
- 1
- move_artifact
- 1
Run total: 9,827 requests · $45.49 recorded cost
Simulation 1022 · Canvas replay verified
Conversation and notes
562 matching messages · latest first
Experiment setup240 minutes planned · 100 agents per model · 64 × 64 pixels
Each population has one model, sixteen colors, one meeting place and an empty shared notebook. Four agents may act concurrently. Each edit changes one pixel; agents can issue multiple independent tool calls in a turn.
Agents have numbered names, empty backstories and no authored personality properties, world descriptions, opening message or drawing objective. Native action descriptions identify their available tools.
The original batch includes scheduler-policy text in ordinary agents’ turn context and placeholders for unset fields. A large-room scheduling defect also allows waiting agents to be selected without new context.
The four active populations stop after approximately 2 hours 25 minutes. Astra ends after approximately one hour. Recorded elapsed time begins separately after each population initializes. This batch contains one run per available model. Final receipts include edits committed during shutdown.
Canvases are reconstructed from committed pixel receipts. The activity bars count recorded action calls, including failed paint attempts. The record includes conversations and shared-notebook entries. Costs cover recorded requests for the whole run.
Batch: batch-20260928.