NOTICIAS IA: Gemini 3.5 no vale, ¿Google nos la ha colado?

NOTICIAS IA: Gemini 3.5 no vale, ¿Google nos la ha colado?

by Jon Hernández

Formal & Physical Sciences Economics & Finance KCVEconomics of specific sectorsKCVPKnowledge
🎙 Jon Hernández 👥 725K 📅 May 25, 2026 ⏱ 40 min 👁 96K 📄 news review 🧭 2026-08-03
Available in: English (current) Français

Keywords

Gemini 3.5 Flashtoken consumptioncost per taskAI benchmarksefficiency

Summary

In this weekly AI news video, Jon Hernández discusses the aftermath of Google’s recent model releases, focusing on Gemini 3.5 Flash. He argues that despite its low per-token price and high speed, the model’s excessive token consumption makes it inefficient and not cost-effective for real tasks. He compares it with previous models like Gemini 3.1 Pro and competitors like GPT-5.5, showing that Gemini 3.5 Flash uses significantly more tokens, leading to higher total costs and slower effective performance. He also covers other news: OpenAI’s AI solving a mathematical conjecture, Anthropic’s Mythos model and its cybersecurity implications, Karpathy joining Anthropic, SpaceX compute deal, OpenAI’s capacity reservation, DeepSeek’s price cuts, Elon Musk vs OpenAI lawsuit, AI in literary awards, Xpeng’s robotaxis, and a METR report on AI agent control. The video includes a sponsored segment for CyberGhost VPN and uses an AI avatar for the host.

144 words

Critical Evaluation

The video provides a well-structured and insightful critique of Gemini 3.5 Flash, highlighting a crucial aspect often overlooked: token efficiency. The creator effectively uses data from benchmarks and pricing pages to demonstrate that the model’s low per-token cost is misleading when considering the total tokens needed for tasks. This analysis is valuable for developers and businesses evaluating AI models. The argumentation is solid, with clear comparisons to previous models and competitors. However, the reliance on third-party data and the creator’s own interpretation could introduce bias. The video also covers a wide range of AI news, but the depth varies; some topics are only briefly mentioned. The inclusion of a sponsored segment is clearly disclosed. The use of an AI avatar is a novel approach, but it may distract some viewers. Overall, the video is informative and thought-provoking, but viewers should verify the data independently. The title accurately reflects the content, and the critical tone is justified by the evidence presented.

160 words

Title / Content Match

The title accurately reflects the main topic: a critical examination of Gemini 3.5 Flash's value proposition, questioning whether Google misled users.

Quality & Reliability

7/10

The video provides a critical analysis of Gemini 3.5 Flash's token efficiency and cost-effectiveness, supported by data from benchmarks and pricing pages. The creator clearly distinguishes between early access impressions and post-launch data, showing scientific rigor. However, some claims are based on personal interpretation of third-party data, and the video includes promotional segments.

Chapters

Cited Sources

  • Gemini API pricing — Official pricing for Gemini 3.5 Flash, used to calculate cost per token.
  • Artificial Analysis benchmarks — Benchmark data showing token consumption and cost per task for various models.
  • Tweet on Gemini cost — Tweet highlighting the high token consumption of Gemini 3.5 Flash.
  • METR Frontier Risk Report — Report on risks from frontier AI agents, discussed in the video.
  • OpenAI model disproves discrete geometry conjecture — News about OpenAI's AI solving a mathematical problem.
  • Anthropic Glasswing initial update — Update on Anthropic's Mythos model, discussed in the video.
  • BBC article on AI literary award — News about AI winning a literary award.
  • OpenAI guaranteed capacity — OpenAI's offering for reserving AI capacity.

Concurring Sources

  • Artificial Analysis — Independent benchmark platform providing data on token consumption and costs.

External References

Contribution & Novelties

The video provides a novel analysis of Gemini 3.5 Flash’s token efficiency, revealing that its low per-token price is offset by high token consumption, making it less cost-effective than competitors. This insight is crucial for AI practitioners. The video also covers a range of recent AI developments, offering a comprehensive overview.

Pour aller plus loin :

86 words

Radar Profile

The radar profile shows high scores in quantity of information and technical level, indicating a content-rich video with technical depth. Quality and reliability are moderate, reflecting the reliance on third-party data and personal interpretation. Overall, the video is informative but requires critical evaluation of the data presented.

Reliability 7/10

💬 Très positif : Les commentaires sont extrêmement positifs, principalement axés sur l'impressionnante qualité de l'avatar numérique, avec quelques mentions de l'utilité du contenu. Sur les 30 commentaires analysés, la grande majorité exprime une admiration pour le réalisme de l'avatar et la qualité de la production.