DeepWism R2 is a novel AGI system designed around the T3CEDS framework, which posits intelligence as entropy reduction rather than attention modeling. Targeting AI researchers and power users, it aims to provide superior performance in complex reasoning and problem-solving tasks by leveraging crowd intelligence mechanisms.

How It Works

DeepWism R2 employs a three-layer architecture (Thin-Thick-Thin Crowd Entropy Dynamics System - T3CEDS). The Thin Perception Layer efficiently processes high-dimensional inputs, reducing entropy. The Thick Processing Layer utilizes crowd intelligence for structured reasoning and collaborative processing to further reduce entropy. The Thin Decision Layer distills complex representations into coherent, low-entropy outputs for actionable decisions. This entropy-centric approach is claimed to be more effective for uncertain problem spaces than traditional attention models.

Quick Start & Requirements

• Access: Interactive chat interface available at i.deepwism.com.
• Open Source: Framework will be fully open-sourced with global API access for authenticated users.
• Contact: For inquiries, contact r2@deepwism.com.

Highlighted Details

• Achieves 27.5% accuracy on Humanity's Last Exam (HLE), surpassing OpenAI DeepResearch (26.6%).
• Claims state-of-the-art performance on xbench-ScienceQA (70.0%) and xbench-DeepSearch (64.0%).
• Reports high accuracy on AIME2025 (93.3%), GPQA-Diamond (88.0%), MMLU (86.0%), and SWE-bench Verified (72.0%).
• Features an interactive chat interface with entropy visualization.

Maintenance & Community

• Contact: r2@deepwism.com
• Website: www.deepwism.com
• Social: Twitter: @DeepWism
• Support: GitHub Issues for bug reports and feature requests.

Licensing & Compatibility

• The framework is slated for full open-sourcing with global API access for authenticated users. Specific license details are not yet provided.

Limitations & Caveats

The project is described as "Research&Report" and "next-generation," suggesting it may be in an early or experimental stage. Specific implementation details and the open-sourcing timeline are not fully elaborated.