🤖 Meta ships Muse Spark 1.3 rivaling Claude and GPT 🤖
Meta's latest model challenges frontier pricing while matching top-tier performance.

Meta has released Muse Spark 1.3, a model it claims rivals Claude Fable 5.1 and surpasses GPT-5.6 Sol, particularly in coding. It uses 25% fewer tokens, handles multiple workflows simultaneously, and asks for confirmation before irreversible actions. Spark 1.3 (Max) scores 62 on the Artificial Analysis Intelligence Index, trailing only Anthropic's top models while costing significantly less — aggressive pricing that could shift how developers choose API providers. Zuckerberg called it "frontier performance almost too cheap to meter" and previewed a larger model codenamed "Watermelon." Available now via Meta's API and Muse Code, with consumer rollout coming soon. Whether Meta will open-source the weights remains undecided.

🔍 Google ships Gemini 3.8 Flash at $0.75 input 🔍
Google keeps Gemini pricing flat while pushing coding performance forward.

Google has released Gemini 3.8 Flash with improved coding, agentic, and multi-step reasoning capabilities, maintaining introductory pricing at $0.75/$3.75 per million tokens. A specialized Flash Cyber variant also launches for vulnerability detection and automated patching, available through a restricted defender program. The model scores 59 on the Artificial Analysis Intelligence Index. Notably, DeepMind's Koray Kavukcuoglu acknowledged that Gemini currently sits "a little below the frontier" — a rare admission that signals Google is prioritizing frequent, cost-efficient updates over headline-grabbing leaps. This pricing pressure matters for every team budgeting API spend.

🛡️ Nvidia and CrowdStrike ship SafeMind AI 🛡️
Nvidia expands beyond chips into enterprise cybersecurity integration.

Nvidia and CrowdStrike have introduced SafeMind, a new family of agentic AI models built specifically for cybersecurity operations. The models can both identify and autonomously close attack paths, combining Nvidia's AI infrastructure with CrowdStrike's security domain expertise. This marks a notable expansion of Nvidia's ambitions beyond hardware into specialized enterprise software — a signal of competitive pressure across the security stack. The collaboration highlights the growing trend of AI-native cybersecurity tools that move beyond detection to active remediation, potentially cutting response times for enterprise customers.

INTERESTING TO KNOW

🎬 Gemini slashes video analysis tokens by 88% 🎬

Google has rolled out agentic video understanding across Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite, delivering a major speed and cost improvement. Rather than processing entire videos at a fixed frame rate, Gemini now intelligently navigates between moments, frames, audio, and transcripts — revisiting sections as needed. Google reports up to 88% fewer tokens, 66% cost reduction, and 7% accuracy gains, a capability jump for any team processing long-form video. Available now via the Gemini API, with broader rollout in the Gemini app and Ask YouTube planned later.

💻 Perplexity ships Hybrid Mode for local Mac AI 💻

Perplexity has unveiled Hybrid Mode for its Computer product, enabling a split architecture where simpler tasks run locally on Mac while harder ones route to the cloud — a rollout designed to keep sensitive data off third-party servers, which matters for enterprise adoption. Three local models are planned: a ~19GB proprietary Perplexity model, Qwen 32B, and a smaller Gemma variant for Macs with 16GB RAM. A dedicated Privacy Gate uses a separate local model to inspect outbound data before cloud transmission, automatically flagging personal information.