Newcourse Alpha - The industry's first cognitive model launched by Nextie

What is Newcourse Alpha

Newcourse Alpha is the industry's first cognitive model launched by Nextie (Nextie), deployable on-device with only 4B parameters. The model is led by Li Di, the "father of Xiaoice", and is based on an open-source reasoning model that decouples knowledge from cognition through reinforcement learning, distilling thinking algorithms. The model achieves effects equivalent to GPT-5.4 in swarm intelligence tasks, with computing costs dropping precipitously, and supports 7×24 active operation. The model is paired with the Tuanzi multi-agent platform, enabling AI to debate, vote, and collaborate on decisions like a team of human experts.

Newcourse Alpha's main features

Cognitive core distillation: Strips away massive factual memory and retains thinking algorithms, teaching the model how to think.
Swarm intelligence collaboration: Provides unified planning and reasoning capabilities for the Harness multi-agent system, supporting multi-agent collaborative thinking.
On-device local deployment: The 4B golden size can run directly on MacBooks, embodied intelligence devices, and other edge devices, without relying on cloud computing power.
Proactive continuous operation: Low-cost support for 7×24 uninterrupted autonomous planning and execution, shifting from passive response (Reactive) to proactive action (Proactive).
Five-dimensional cognitive evaluation: Quantifies swarm intelligence quality based on perspective completeness, implicit demand satisfaction, dialectical depth, practical feasibility, and decision explainability.

Newcourse Alpha's technical principles

Decoupling knowledge from cognition: Nextie chose to perform reinforcement learning on top of an existing open-source reasoning model. By decoupling knowledge from cognition, it distills the model's cognitive core, focusing training objectives on generalization and abstraction capabilities, enabling the model to draw inferences in different scenarios.
220 years of swarm intelligence evolution training: The team reviewed human academic papers spanning 220 years from 1800 to 2020, summarizing the evolutionary trajectory of swarm intelligence. It deeply deconstructs the mechanisms by which humans make better decisions through debate, reflection, challenge, and voting, translating them into a machine-learnable cognitive framework that provides theoretical support for the cognitive model.
Small parameters + high-quality architecture: Continues the Xiaoice team's small-parameter + high-quality architecture route. The 4B parameter count has been verified as the golden size, large enough to carry complex thinking algorithms and small enough to run on-device. Cognitive architecture, rather than parameter count, is regarded as the long-term barrier.

How to use Newcourse Alpha

Visit the platform: Open the Tuanzi multi-agent platform beta address in a browser https://mytuanzi.com/ to experience swarm intelligence collaboration.
Register and log in: During the beta period, a "power" mechanism is used instead of computing power credits.
Choose a scenario: Select preset agent combinations such as Sister Team, Research Team, or Ask Miracle based on your needs.
Launch a task: Enter a question or requirement, and the system automatically dispatches multiple AI Agents with different expertise to collaborate.
View decisions: Observe the debate, challenge, reflection, and voting process among multiple Agents, and ultimately obtain the optimal solution output after collective deliberation.

Newcourse Alpha's core advantages

Achieving more with less: The model has 4B parameters and achieves output effects equivalent to GPT-5.4 in swarm intelligence tasks, breaking the Scaling dilemma that parameters are justice.
Precipitous cost reduction: runs on-device, eliminating cloud deployment and token billing; computing cost drops by an order of magnitude, from burning GPUs to paying electricity bills.
Generalizable cognition: strips away factual memory and retains thinking algorithms, enabling the transfer of thinking strategies from a single scenario to unrelated domains and achieving cross-domain extrapolation.
Transparent and traceable process: continues the Xiaoice Chain approach; the Agent's thinking process is observable and verifiable, providing a trustworthy foundation for multi-agent collaboration.
Proactive continuous operation: low power consumption supports 7×24 uninterrupted autonomous planning and execution, shifting the Agent from passive response to proactive action.

Comparison of Nextie Alpha with similar competitors

Comparison dimension	Nextie Alpha (Nextie)	DIKWP cognitive model (Duan Yucong's team)
Core positioning	Lightweight cognitive core engine (4B parameters)	Five-layer cognitive architecture theoretical framework (DIKWPaaS platform)
Cognitive architecture	Strips away factual memory and retains thinking algorithms, strengthening generalization and abstraction capabilities	Data→Information→Knowledge→Wisdom→Purpose five-layer networked semantic structure, emphasizing "Purpose" guidance
Parameter scale	4B (deployable on-device)	Not bound to a specific model; can be layered on top of any large model
Training method	Reinforcement learning based on open-source reasoning models, decoupling knowledge and cognition	Formally defines reasoning rules through "semantic mathematics", realizing formalization of the cognitive process
Evaluation system	Five-dimensional swarm intelligence evaluation: perspective completeness, implicit demands, dialectical depth, practicality, interpretability	Artificial consciousness "white-box" assessment, analyzing data/information/knowledge/wisdom/purpose capabilities layer by layer
Deployment form	On-device model + Tuanzi multi-Agent platform (productized)	DIKWPaaS semantic platform (academic/platformized), emphasizing "semantics as a service"
Human role	Humans participate in multi-Agent debate and voting, and can intervene in decisions in real time	Human purpose serves as the highest layer (Purpose), guiding the entire cognitive process
Typical scenarios	Complex business decisions, entrepreneurial consulting, risk assessment, proactive Agents	Proactive AI, proactive medicine, artificial consciousness research, semantic closed-loop reasoning
Transparency	Thinking process is observable and traceable (continuing the Xiaoice Chain approach)	Emphasizes "white-box" analysis; each layer of cognitive capability can be independently audited and assessed
Cost structure	Low power consumption on-device, sustainable 7×24 operation	Depends on cloud-based distributed architecture; cost is tied to the chosen underlying model

Application scenarios of Nextie Alpha

Complex business decisions: corporate strategy formulation and investment risk assessment, using multi-Agent perspective completeness to identify potential loopholes and external risks.
In-depth research and analysis: academic research and market research, breaking through the cognitive limitations of a single model through cognitive collisions among interdisciplinary expert teams.
Smart home robots: On-device deployment gives home robots continuous autonomous planning capabilities, with controllable computing costs, making everyday scenarios economically viable.
Entrepreneurship Consulting and Coaching: Simulating MiraclePlus scenarios, multiple Agents challenge the business plan from different dimensions, seeking the optimal solution rather than a standard answer.
Content Safety Review: As in the Guardian Shrimp case, swarm intelligence can accurately identify hidden high-risk vulnerabilities and adjust output in real time.