Perplexity Launches AI Computer as Anthropic Scraps Safety Pledge Amid Pentagon Tensions
New Products & Services
Perplexity Launches "Perplexity Computer": Perplexity AI has introduced an integrated system capable of handling end-to-end projects including research, design, coding, and deployment. Initially available to Max users, the service utilizes a usage-based pricing model, which the company advocates as a more sustainable alternative to traditional subscriptions.
AI Models & Benchmarks
Performance Evaluations for Gemini 3.1 and Qwen 3.5: Recent benchmark results show Gemini 3.1 performing well in reasoning and frontend development, though users raised concerns regarding potential benchmark manipulation. Meanwhile, Qwen 3.5 showed significant limitations on complex, real-world coding repositories when compared to models like GPT-5.3 Codex.
- Gemini 3.1 livebench results
- Qwen 3.5 craters on hard coding tasks — tested all Qwen3.5 models (And Codex 5.3) on 70 real repos so you don't have to.
AI Development & Programming
AI Coding Agents Reach New Autonomy Milestone: Andrej Karpathy reports that programming practices have shifted drastically in early 2026, with AI agents becoming reliable enough to handle most coding tasks autonomously. Developers are increasingly moving into oversight roles, focusing on reviewing and debugging code generated by advanced models like Claude and ChatGPT 5.1 Codex.
AI Safety, Ethics & Policy
Anthropic Faces Policy Criticism and Government Friction: Anthropic has sparked controversy by dropping its flagship safety pledge and facing potential blacklisting by the Pentagon for refusing to comply with military requests. In a related community backlash, a developer open-sourced 155,000 Claude conversations and released a tool called "DataClaw" to protest the company's restrictive data policies.