Claude Sonnet 4.5 is now the best coding model

Thinking Machines Lab

Tinker democratizes language model fine-tuning by offering researchers a managed, cost-effective API that handles infrastructure complexity while maintaining control over training methods and data. This enables broader participation in cutting-edge model customization without requiring in-house distributed training expertise.

Alephic

Claude Code's breakthrough success comes from combining filesystem access with Unix command integration, enabling stateful AI agents that overcome context window and memory limitations. This approach demonstrates a powerful template for building reliable agentic systems around existing model capabilities rather than constraining them.

The Zvi

Claude Sonnet 4.5 represents a significant leap in coding and agentic capabilities with state-of-the-art benchmarks and improved system prompt, positioning Anthropic competitively against OpenAI's GPT-5 despite higher per-token costs for non-subscription users.

TechCrunch

OpenAI's launch of Sora, an AI video social media app, has created internal friction between researchers worried about mission drift and leadership defending the consumer business as necessary funding for AGI research. The move raises questions about whether profit incentives will ultimately override the company's nonprofit charter.

Google Developers Blog

Google released Tunix, a JAX-native library that simplifies LLM post-training with modular APIs for fine-tuning, preference optimization, reinforcement learning, and distillation, enabling faster iteration on model alignment for production deployment.