subora.com

What It Is

Subora describes itself on its page as a “General Reasoning Model.” Its core claim is to achieve “true grounding, no memorization” — in other words, emphasizing genuinely grounded reasoning rather than relying on memorized training data. Using the ARC-AGI-2/3 narrative, the page argues that today’s frontier AI systems consume enormous amounts of energy yet still struggle to match human-level reasoning on novel tasks.

Core Capabilities and Use Cases

Based on the captured page content, Subora does not appear to offer a directly usable chatbot, Agent, API, or enterprise tool. It looks more like a research or concept page focused on the direction of general-purpose reasoning models. Rather than listing product features, it criticizes the performance of existing large models on ARC-AGI-2: before the “leak,” models such as GPT-4o, Claude 3.7, DeepSeek R1, and o3 scored very low; after the “leak,” newer models achieved higher scores, but the page argues that this still reflects memorization loops rather than true reasoning. Its likely audience is AGI researchers, people tracking model evaluations, and those interested in changes to AI architectures.

Pricing, Trials, and Integration

The page does not provide free quotas, trial access, pricing plans, API documentation, SDKs, enterprise integration options, or payment information. It also does not state whether registration is currently open. As a result, it is not yet possible to determine whether this is a purchasable product, a closed research project, or simply a conceptual showcase.

Pros and Cons

Its main strength is its highly focused positioning: it makes a clear argument around “general reasoning” and “non-memorization,” reinforced by benchmarks such as ARC-AGI. It may be useful for people following AGI evaluation and debates around model architecture. The drawbacks are also obvious: there are no verifiable benchmark results for Subora’s own model, no output examples, no technical white paper, no privacy policy, and no product roadmap. The page makes strong claims about future models and data leakage, but users cannot assess practical usability based on the information provided.

Who It’s For and Access from China

At this stage, Subora is better suited for researchers, AI investors and industry analysts, and AGI enthusiasts to monitor, rather than as a mature productivity tool to procure. The page does not mention access from China, and network connectivity, payment methods, and local compliance information are all unknown. If you need something immediately usable, more mature platforms such as OpenAI, Claude, Gemini, DeepSeek, and Mistral are worth considering.