1× screenshot replay, without audio. The last actual observation stays on screen until the next one; a still frame does not mean the task paused.
Full task instruction
SHA-256
REALTIME GUI · FIXED COHORT / 13F1B74
Compare GPT-6.1 Sol and Claude Haiku 5.5, then watch what happened in each task.
SAMPLED PERFORMANCE
One valid run per model per task. Scores are task points, not success rates. The aggregate is the equally weighted mean over these 17 task variants.
INSPECT THE EPISODES
| Task | GPT | Haiku |
|---|
Select a score to watch that model’s run.
1× screenshot replay, without audio. The last actual observation stays on screen until the next one; a still frame does not mean the task paused.
OTHER TASKS & BUDGETS
Existing Blender and TicketVault pilots use different budgets and protocols. Watch them here; their scores are separate from the leaderboard above.