Claude 4.5 & GLM-4.6 Clash as Open-Source 1T Model Rings in New AI Era
New AI Model Releases
Anthropic Releases Claude 4.5 Sonnet: Anthropic’s latest model, Claude 4.5 Sonnet, demonstrates significant improvements in agentic coding, tool use, and reasoning, outperforming competitors like GPT-5 and Gemini 2.5 Pro in multiple benchmarks.
GLM-4.6 Launches, Claiming Superiority Over Claude 4.5: Z.ai’s GLM-4.6 is released with benchmarks suggesting it outperforms Claude 4.5 Sonnet, offering advanced reasoning and performance.
Ring-1T-Preview: 1T Open-Source Reasoning Model: A new open-source model with 1 trillion parameters and 50B activations, Ring-1T-preview, is released, pushing the boundaries of open-source AI capabilities.
AI Model Efficiency & Cost Innovations
LoRA Matches Full Fine-Tuning Performance with Fewer Resources: A blog post by Thinking Machines reveals that LoRA (Low-Rank Adaptation) in reinforcement learning can achieve results comparable to full fine-tuning while using fewer computational resources.
DeepSeek V3.2’s Cost Efficiency Explained: DeepSeek V3.2 leverages a near-linear architecture to reduce attention complexity, making it significantly cheaper than competing models.
AI Platform Integrations
Claude 4.5 Sonnet & 4.5 Thinking Available on Perplexity: Perplexity Pro and Max subscribers now have access to Claude 4.5 Sonnet and 4.5 Thinking, enhancing the platform’s AI-driven search and reasoning capabilities.