Baidu ERNIE multimodal AI beats GPT and Gemini in benchmarks

Plus: Google reveals its own version of Apple’s AI cloud

Author

Hafizur Rahman
November 13, 2025

In partnership with

Today's Newsletter Highlights:

AI NEWS TODAY

Shoppers are adding to cart for the holidays

Over the next year, Roku predicts that 100% of the streaming audience will see ads. For growth marketers in 2026, CTV will remain an important “safe space” as AI creates widespread disruption in the search and social channels. Plus, easier access to self-serve CTV ad buying tools and targeting options will lead to a surge in locally-targeted streaming campaigns.

Read our guide to find out why growth marketers should make sure CTV is part of their 2026 media mix.

Baidu’s ERNIE-4.5-VL-28B-A3B-Thinking is a new multimodal AI model. It can manage dense, non-text inputs, such as technical diagrams, factory video feeds, and logistics dashboards.

The model is “lightweight” in inference. It activates only 3 billion parameters during operation, while having a total of 28 billion parameters.

Key Points:

  • In key benchmark tests, Baidu claims ERNIE outperforms GPT-5-High and Gemini 2.5 Pro.

  • ERNIE focuses on tool use and “action” instead of just perception. It can zoom into photos, start image searches for unknown objects, and give structured data like JSON with object coordinates.

  • Many business cases, like engineering schematics and manufacturing videos, need multimodal reasoning. Baidu is targeting these areas.

  • Gains: Activating only 3 billion parameters lowers inference costs and reduces computing needs. This makes deployment easier for larger scales.

  • Competitive pressure: ERNIE appears to outperform GPT and Gemini on some benchmarks. This hints at a change in the “few large models dominate” era. Other players can now compete.

  • Broader implications: If businesses can access better multimodal AI at a lower cost, it could boost AI use in manufacturing, logistics, and professional services, not just for web and text tasks.

Master Marketing Measurement with Incrementality

Still guessing which campaigns actually work? Discover how leading brands cut waste and scale smarter with incrementality. This free guide breaks down how to measure true impact, optimize spend, and move beyond outdated attribution. Download now and start making smarter, data-backed marketing decisions.

Google reveals its own version of Apple’s AI cloud

Google has launched a platform called Private AI Compute. This cloud-based system uses advanced AI processing. It keeps your data private, just like on-device models do.

The system uses Google’s best models, like the Gemini models, within a secure environment. This setup ensures that no one, including Google, can access your data.

Key Points:

  • It has a special hardware and software setup, including Google’s TPU infrastructure, within secure enclaves.

  • Key security measures include end-to-end encryption, remote attestation, and a “zero-access” design. This setup keeps users’ data under their control.

  • Google points out some use cases. This includes better tips for devices like Pixel phones. It also offers improved transcription and summarisation in different languages.

  • Google sees this as “just the beginning.” This move shows a shift where cloud AI must provide high computing power while ensuring privacy and trust.

Microsoft's Impressive Free Cash Flow - MSFT Stock Could Be Worth 28% More

Veir, backed by Microsoft, is making superconducting cables for data centres and AI racks.

Their first product is a low-voltage cable system that carries 3 megawatts in one run.

This system uses high-temperature superconductors cooled by liquid nitrogen (around -196 °C). It aims to cut down on space and heat problems linked to traditional copper feeders in dense AI environments.

Key Points:

  • Veir has built a simulated data centre near its Massachusetts HQ to test this setup and plans to launch commercially by 2027.

  • Power density challenge: AI racks require more power than standard setups. Veir’s technology solves the issue of delivering power in tight spaces with minimal loss.

  • By replacing large copper feeders with superconducting cables, data centres can save space. They also reduce heat output and simplify power routing.

  • Infrastructure for the next wave: As AI and high-performance computing expand, key support systems—like power, cooling, and layout—become vital. This strategy positions Veir and its backers ahead of competitors.

  • Green / sustainability angle: Reduced resistive loss means less wasted energy. Smaller routing and cooling needs could lead to lower OPEX for operators.

TOP NEW AI TOOLS

Free AI Tools You Shouldn’t Miss

🔥 Talo - All‑in‑one AI translator: calls, events, streaming & API!

🔥 TrustMRR - The database of verified startup revenues!

🔥 Emma - AI Nutrition intelligence that understands food globally!

🔥 YouArt - An agentic workflow studio to create high-quality creatives!

Move us from the 'Promotions' to your 'Primary Inbox' and get AI news, tips, and tutorials delivered straight to you. Don't miss your chance to stay on top of the AI industry's latest buzz!