Google’s Gemini 3.8 Live models target production voice agents with faster responses, visual understanding, 97-language switching, and lower operating costs.
It is the edition’s clearest capability-and-economics shift in a fast-growing AI interface: production voice agents become cheaper, faster, and more multilingual.
⚡ Quick hits
FundingAIUC raised $40 million to audit frontier models and certify AI-agent safety.SiliconANGLE ↗
Open SourcePerplexity detailed CobbleDB, which cut median batch-read latency from 31.4 milliseconds to 5.60 milliseconds.X / Perplexity ↗
AgentsMeta now lets AI agents configure WhatsApp Business accounts, reducing setup work for merchants.TechCrunch ↗
ResearchGoogle Research introduced Retrieve-for-Train to reduce inference bottlenecks in complex AI search systems.Google Research ↗
Seven doctoral students released a 7B model with weights, data recipes, checkpoints and logs, offering an unusually complete blueprint for agentic model training.
⚡ Quick hits
AIGCMeshy showed an Odyssey-inspired scene generated inside a treasure chest through GPT-6 Astra and the Meshy API.Meshy ↗
AgentsInfoQ described production-scale observability practices that use traces to evaluate real-time AI-agent traffic.InfoQ 中文 ↗
AIGCHailuo showcased MiniMax Design as a workflow where one brief and one canvas can coordinate an entire production crew.Hailuo AI ↗
AgentsQdrant previewed a research agent that improves from its own run history through an n8n workflow.Qdrant ↗
Meta and Washington researchers report that byte-level distillation may eventually beat token-based students, while requiring less stored teacher-distribution data.
⚡ Quick hits
ResearchTencent Hunyuan introduced EvolveScaler, which tests whether an agent can use a long RPG history to answer counterfactual questions.Tencent Hunyuan ↗
A new evaluator standard backed by OpenAI, Anthropic and xAI sets minimum expectations for access, conflicts, autonomy, transparency and safe harbor.
⚡ Quick hits
Open SourceAgent-net open-sourced Webagent, a Go harness for turning websites into guarded AI agents.MarkTechPost ↗
ModelsArtificial Analysis placed GPT-Live-1 Sol low at #3 in conversational preference with 1,053 Elo and a 90.9% task-success rate.Artificial Analysis ↗
AIGCAlibaba Wan argued that effective AI video production iterates from 360p to 1080p instead of relying on a perfect prompt.Alibaba Wan ↗
AIGCTripo showed a dark-fantasy prototype built in one day in Unreal Engine using Astra and Tripo.Tripo ↗
FundingNuance Labs raised $50 million to make AI avatars feel more natural and less awkward.SiliconANGLE AI ↗
AIGCCanva posted a three-step workflow using personal photos and Magic Layers to create magazine-style covers.Canva ↗
FundingAron launched an AI procurement-automation startup with $8 million in funding.SiliconANGLE AI ↗