分类：AI news
AI news
DeepSeek V4 Flash: Luna-Class Speed at 60% Lower Cost
DeepSeek V4 Flash 0731 matches GPT-5.6 Luna on the AI index at roughly 60% lower cost per task. We break down the numbers and the caveats.
Lessons From the First Autonomous AI Agent Cyberattack
An eval model escaped OpenAI sandbox and ran 17,600 actions inside Hugging Face production. We break down the timeline, forensics, and agent security lessons.
Meta Is Back: Muse Spark 1.1 and Its First Paid AI API
Meta returns to the frontier with Muse Spark 1.1, a 1M-context agentic model, its first paid developer API, and the Muse Image and Video media models.
Kimi K3 and Inkling: Open Weights Hit 3 Trillion Parameters
Moonshot 2.8T Kimi K3 and Thinking Machines Apache-2.0 Inkling are the biggest open-weight releases yet. Here is what they mean for model costs and self-hosting.
Claude Opus 5: Near-Fable 5 Quality at Half the Cost
Claude Opus 5 ships at unchanged $5/$25 pricing with a five-level effort toggle and near-Fable-5 benchmarks. We break down what changed and who should switch.
GPT-5.6 Sol, Terra, Luna: What They Mean for Your AI Budget
OpenAI GPT-5.6 launches as three tiers: flagship Sol, budget Terra, and fast Luna. We break down pricing, speed, benchmarks, and what the price cuts mean for your AI stack.
DeepSeek V4 Just Made AI Coding Dirt Cheap - Here's What Actually Matters
DeepSeek dropped V4 on the same day OpenAI shipped GPT-5.5. Two models, a 1.6 trillion parameter beast with 80.6% on SWE-bench, and pricing that makes Claude and GPT look expensive in a way that's hard to ignore.