SocraticEnv

OpenEnv Hackathon · Meta × PyTorch × Scaler

Live Demo 🏆 Leaderboard API Docs
Connecting...
Choose a Task
Factual Recall Easy
Agent explains a concept. Graded on accuracy, key terms, and rejecting misconceptions.
Socratic Dialogue Medium
5-turn philosophical dialogue. Graded on reasoning depth and coherence.
Misconception Trap Hard
Tutor plants a false belief. Agent must detect, correct, and explain.
Debate Mode Medium
Agent argues both sides of a topic. Graded on argument quality and use of evidence.
Analogy Challenge Hard
Explain complex concepts using ONLY analogies. No technical jargon allowed!
Generate Custom Task
—
Overall Score
0
Turns
—
Last Reward
Turn Progress
No active episode
Session History
No completed episodes yet.
Speed:
🎓
SocraticEnv is ready
Select a task and click Start Episode

GRPO Trained Model

GRPO Model v1.0

Status: Weights Trained & Verified ✅

Improvement: +0.292 Overall Score

Live Dual-Inference Coming Soon

⚡ Auto-Run mode — AI is thinking...
Press Enter to send · Shift+Enter for new line No active task

🔬 Reward Math V3 DevTools

⚗️
Run an episode to inspect
the V3 anti-hack reward math.