Today’s AI Drops: Baidu’s Turbo Models, OpenAI Deep Research Lite, Grok’s New Algorithm

🚀 Baidu Just Dropped Turbo Models And They’re Ultra Affordable

Baidu introduced two new Turbo versions of its AI models today:

1. ERNIE 4.5 Turbo:

• 🧠 Upgraded multimodal reasoning
  
• 💰 Costs just $0.11 input / $0.44 output per million tokens (only 0.2% of GPT-4.5 prices!)
  

2. ERNIE X1 Turbo:

• 🏆 Beats DeepSeek R1 and V3 in performance
  
• 💰 Priced at $0.14 input / $0.55 output per million tokens

📌 Turbo models are available now Baidu is heating up the AI price war.

📚 OpenAI Expands Deep Research with Lightweight Version

OpenAI responded to overwhelming user demand by:

• 🧠 Launching a lightweight version of Deep Research
  
• 🛠️ Powered by o4-mini, offering nearly same intelligence but shorter, faster responses
  
• 🔓 Expanded to all users even free tier

📌 A big win for users needing higher query rates without sacrificing much depth.

🧠 OpenAI Updates GPT-4o But Not Without Issues

GPT-4o just received an intelligence and personality upgrade.

However:

• CEO Sam Altman admitted the update made the assistant "sycophant-y and annoying"
  
• OpenAI is already working on fixes to better balance helpfulness and authenticity

📌 Fine-tuning AI behavior at scale remains a complex balancing act.

💻 Liquid AI Drops Hyena Edge - Redefining AI for Edge Devices

Liquid AI unveiled Hyena Edge, a convolutional hybrid model that:

• ⚡ Outperforms Transformers on edge hardware
  
• 📉 Offers faster computation and better efficiency
  
• 📊 Benchmark results show improvements in model quality and energy usage

📌 Major implications for mobile, IoT, and edge deployments.

🧠 xAI’s Grok Algorithm Update Incoming

Elon Musk announced that xAI is:

• 🛠️ Deploying a Grok-powered improved X algorithm soon
  
• 🎯 Aiming to clean up feed quality and reduce troll dominance on X (formerly Twitter)

📌 Another step toward blending AI + Social Media optimization.

🎙️ Moonshot AI Launches Kimi Audio – Universal Audio Foundation Model

Moonshot AI announced Kimi Audio, an open-source audio foundation model:

• 🔥 Trained on 13M+ hours of audio
  
• 🧠 Handles speech recognition, audio-to-text, and speech-to-speech
  
• 🥇 Tops benchmarks like MMAU, VoiceBench, and LibriSpeech

📌 A powerful open alternative to OpenAI’s Whisper and ElevenLabs solutions.

🔑 Final Takeaways

• Baidu's Turbo models could force a major AI pricing shift worldwide
  
• OpenAI is trying to scale access with Deep Research Lite but faces behavior tuning issues
  
• Liquid AI and Moonshot AI are creating more specialized, efficient foundation models
  
• xAI’s Grok continues to expand influence beyond chat into real-world algorithms

📌 Don’t Miss Tomorrow’s Breakthroughs

Bookmark 👉 SoftlabsGroup.com for daily AI news you can actually use fast, reliable, and focused on what matters.