We’ve got a new king of the 1B models!
LLMS
-
Claude 2.5 Implicit Caching and AI Studio Updates
By
–
Ack, I am on it. – New implicit caching feature for 2.5 models is landing next week (no need to explicit cache unless you want to guarantee cache hit)
– Will follow up on logic for explicit caching (timeouts)
– Will see if we can return cache hits more clearly – New AI Studio -
AI’s First Decade: Methods Over Benchmarks, RL’s Missing Piece
By
–
Beautifully written post from Shunyu connects the dots the past 10 years in AI:
– Winners of the “first half” of the AI game have been methods papers, not benchmarks
– It turned out the missing piece in getting RL to work was priors (natural language reasoning), obtained in a way -

Claude.ai Now Supports Custom MCP Servers via Remote URLs
By
–
Not sure if some folks are fully realizing what we launched today so to make it more explicit: You can bring any custom MCP server into claude dot ai now. All you need is to host it somewhere (remote MCP) and provide the URL link to add it in.
-
Qwen3 32B Achieves Record 282 Tokens Per Second
By
–
So fast it's almost unbearable 🐻
— SambaNova (@SambaNovaAI) 1 mai 2025
Our recently launched @Alibaba_Qwen's Qwen3 32B on SambaNova Cloud is officially the fastest!
🚀 282 t/s — verified by @ArtificialAnlys!
🌍 Multilingual mastery
🧠 Smarter reasoning
📈 Built for complex agent appsSo fast it's almost unbearable Our recently launched @Alibaba_Qwen
's Qwen3 32B on SambaNova Cloud is officially the fastest! 282 t/s — verified by @ArtificialAnlys
! Multilingual mastery Smarter reasoning Built for complex agent apps -

AI Model Fatigue: Beyond the Hype Cycle
By
–
I think I’m over the “new AI model” headlines. I see them day after day. Eyes totally glaze over. One’s cheaper, one’s faster, one’s a little bit better at code, I get it. Even yesterday, my old buddy GPT-4 was removed forever—I felt nothing. I don’t think we’ve hit a ceiling
-

New PHI4 Model Appears on Hugging Face with Feature Summary
By
–
Nouveau Modèle "PHI4" qui vient de pop sur hugginface. Je vous ai résumé ses caractéristiques dans ce tableau
-
Claude Research Mode: Comprehensive Investigations with Integration Support
By
–
2/ When Research mode is toggled on, Claude conducts deeper investigations across hundreds of sources and delivers a comprehensive report on your query. Research works with Integrations as well, allowing Claude to search through data from any of the apps or tools you connect.
-
Claude Integrations: Connect Any Data Source or Tool
By
–
1/ You can now connect any data source or tool to Claude with Integrations.
— Alex Albert (@alexalbert__) 1 mai 2025
Choose from pre-built integrations or build your own MCP server to connect Claude with any tool in <30 minutes.pic.twitter.com/Zz80p9eZWP1/ You can now connect any data source or tool to Claude with Integrations. Choose from pre-built integrations or build your own MCP server to connect Claude with any tool in <30 minutes.
-
Two Major Updates to Claude.ai: Remote MCP Integrations and Advanced Research
By
–
Two big updates to claude dot ai today:
1. Integrations using remote MCP is finally here
2. A more advanced version of Research is out Let's walk through them:
