Une nouvelle IA surpasse les meilleurs modèles… mais personne ne sait qui l’a créée

Une nouvelle IA surpasse les meilleurs modèles… mais personne ne sait qui l’a créée

A new AI surpasses the best models... but nobody knows who created it

🎙 AI Revolution en Français 👥 8K 📅 August 24, 2026 ⏱ 17 min 👁 3K 📄 news review 🧭 2026-09-07
Available in: English (current) Français

Keywords

OxAlphaOpenRoutermodèle furtifGLMbenchmark

Summary

The video investigates the mysterious AI model ‘OxAlpha’ that appeared on OpenRouter, outperforming leading models on a coding benchmark. The creator is unknown, sparking a community investigation. The video presents evidence from tokenizer analysis, video encoder behavior, and benchmark scores, suggesting it might be an unreleased model from Zhipu (GLM family). It also discusses the implications for AI development and the trend of ‘stealth’ model releases. Additionally, the video covers recent developments from Anthropic (Claude Mythos for cybersecurity) and OpenAI (open-sourcing the Codex runtime). The video includes a promotional segment for an investment platform.

94 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides a valuable and detailed analysis of the OxAlpha mystery, presenting a structured investigation with specific technical evidence. The argumentation is solid, as it systematically examines various clues (tokenizer, video encoder, refusal patterns) and compares them with known models. It also contextualizes the benchmark scores, noting the small sample size and differences in evaluation setups. The video effectively balances technical depth with accessibility, making it informative for both developers and AI enthusiasts. However, the evidence is circumstantial, and the conclusion (99% likely GLM) is based on inference rather than confirmation.

Scientific Rigor, Source Quality, Title Accuracy

The video demonstrates a good level of scientific rigor by referencing specific technical details and community investigations. It cites sources like Ben Davis’s benchmark tests, OpenRouter tracking, and community analyses. The title accurately reflects the content. However, the video includes a promotional segment for an investment platform, which is clearly separated but may affect perceived objectivity. The sources are not formally cited with links, but the video mentions names and platforms that can be looked up. Overall, the rigor is acceptable for a news review format.

193 words

Title / Content Match

The title accurately reflects the content: the video discusses a new AI model that outperforms existing ones and the mystery surrounding its creator.

Quality & Reliability

7/10

The video provides a detailed and structured investigation into the mysterious AI model OxAlpha, citing specific technical evidence (tokenizer matches, video encoder analysis, benchmark scores) and referencing community investigations. However, the evidence is largely based on indirect inferences and unverified claims, and the video includes a promotional segment for an investment platform, which may introduce bias. The overall reliability is moderate to good, with a score of 7.

Key Moments

Cited Sources

Concurring Sources

  • OpenRouter — Platform where OxAlpha was released
  • GLM-5 — Suspected model family

Dissenting Sources

  • WCCFTECH article — Suggested Microsoft as the creator, conflicting with the GLM theory

Contribution & Novelties

The video provides an original and timely investigation into the OxAlpha mystery, synthesizing community findings and technical evidence. It offers a clear analysis of the clues and their implications for the AI landscape. The video also highlights the trend of stealth model releases and their strategic value.

Pour aller plus loin :

  • OpenRouter — Platform where OxAlpha appeared.
  • GLM-5 — Zhipu’s model family, suspected to be behind OxAlpha.
  • SWE-bench — Benchmark used to evaluate coding agents.
  • Mixture of Experts — Architecture used by OxAlpha.

84 words

Radar Profile

The radar profile shows high scores in quantity of information and technical level, indicating a detailed and technical review. Quality and reliability are slightly lower, reflecting the speculative nature of the investigation. Overall, the video is informative but relies on indirect evidence.

Reliability 7/10

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