This New AI Beats the Best Models... But No One Knows Who Built It

Item: This New AI Beats the Best Models... But No One Knows Who Built It
Rating: 4
Author: PostSingularity Institute

by AI Revolution

Applied Sciences & Engineering AI — News & Applications UYComputer science UYQArtificial intelligence (AI)

🎙 AI Revolution 👥 566K 📅 August 24, 2026 ⏱ 16 min 👁 38K 📄 news review 🧭 2026-09-07

Available in: English (current) Français

Keywords

Ox AlphaOpenRouterGLMAI benchmarkAI mystery

Summary

The video reports on the sudden appearance of a frontier AI model named ‘Ox Alpha’ on OpenRouter, offering free access with a 1M-token context window and claimed capacity for 100 trillion tokens per day. It quickly outperformed established models like GPT-5.6 Sol and Claude Fable 5 on a coding benchmark (DeepSeek-SWE), sparking a forensic investigation into its origins. Technical fingerprinting, including video token analysis, tokenizer matching, and stylistic traits, strongly suggests it is an unreleased model from Zhipu AI’s GLM family, though no official claim has been made. The video also covers Anthropic’s expansion of Claude Mythos 5 into security tools and a $35M fund for open-source cyber defense, and OpenAI’s open-sourcing of the Codex harness for embedding agents into applications. The narrative highlights the growing difficulty of tracking AI development and the implications for security and software engineering.

139 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides substantial value by aggregating and analyzing technical evidence about a mysterious AI model, offering a balanced view of its capabilities and limitations. The argumentation is structured and evidence-based, presenting multiple hypotheses and weighing them against technical details. However, the inclusion of a promotional segment and speculative theories slightly undermines the objectivity, though the core analysis remains solid.

Scientific Rigor, Source Quality, Title Accuracy

The video demonstrates a good level of scientific rigor by citing specific technical analyses and linking to primary sources in the description. The sources include reputable outlets like Business Insider and official blogs from Anthropic and OpenAI. The title accurately reflects the content, and the video maintains a clear focus on the mystery and its implications. The analysis of the benchmark results is nuanced, acknowledging the limitations of small sample sizes and different benchmark calibrations.

150 words

Title / Content Match

The title accurately reflects the content, focusing on the mystery and performance of the Ox Alpha model.

Quality & Reliability

7/10

The video provides a detailed and nuanced analysis of a mysterious AI model, citing specific technical evidence and multiple sources. However, it includes a promotional segment and relies on unverified claims and speculative theories, which slightly reduce its overall reliability.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

Business Insider: Ox Alpha AI model mystery — Source for the initial report on Ox Alpha's appearance and mystery.
Implicator: Ox Alpha Zhipu GLM tokenizer match — Technical analysis linking Ox Alpha to Zhipu's GLM family.
Anthropic: Bringing Claude Mythos 5 to more defenders — Official announcement about Claude Mythos 5 in security tools.
OpenAI: Unlocking the Codex harness — Official announcement about open-sourcing the Codex harness.

Concurring Sources

Business Insider: Ox Alpha AI model mystery — Reports on the mystery and initial performance, consistent with the video's narrative.
Implicator: Ox Alpha Zhipu GLM tokenizer match — Technical analysis supporting the GLM hypothesis, aligning with the video's evidence.

Dissenting Sources

WCCFTech analysis — The video mentions a competing analysis pointing to Microsoft's MAI family, which contradicts the GLM hypothesis.

External References

go.aiskool.io

Contribution & Novelties

The video provides a unique, real-time analysis of an emerging AI model, combining technical fingerprinting with industry context. It highlights the growing trend of anonymous AI releases and the challenges of tracking frontier AI development.

Pour aller plus loin :

OpenRouter — Platform where Ox Alpha appeared, relevant for understanding the distribution of AI models.
GLM-130B — Background on Zhipu’s GLM model family, which is suspected to be behind Ox Alpha.
SWE-bench — Benchmark used for evaluating AI coding agents, relevant to the performance claims.
Claude — Anthropic’s AI assistant, relevant to the discussion of Claude Mythos 5.
Codex — OpenAI’s coding agent, relevant to the open-sourcing of the harness.

110 words

Radar Profile

The radar profile shows a balanced performance across all dimensions, with slightly higher scores in information quantity and technical level, reflecting the video's detailed analysis and technical depth. The lower score in information quality is due to the inclusion of promotional content and speculative elements.

Reliability 7/10

💬 Sur les 0 commentaires analysés, aucune tendance n'est disponible.