Reader-focused coverage of AI models, APIs, infrastructure, developer tools and changes that matter in practice. Client service announcements remain inside the authenticated account area.
Google Cloud’s Gemini at Work 2026 announcement moves Gemini beyond a single-model assistant, positioning it as an enterprise agent layer with multi-model orchestration, audit controls, sandboxing, and real-time spend caps.
Read news →Microsoft Decision-1 is now available through Vercel AI Gateway, giving developers another typed-decision model for classification, routing, and scoring calls that do not need a full generative response.
Read news →Anthropic has put dynamic workflows for Claude Managed Agents into public beta, letting a managed agent generate and run multi-agent workflow programs in the background.
Read news →OpenAI has released the Decisions API in beta for low-latency typed answers on GPT-6 Luna, creating a cheaper, narrower path for routing, classification, policy checks, and other predefined decision workloads.
Read news →OpenAI has added an Ultrafast service tier for GPT-6.1 Sol in the API, giving latency-sensitive teams a faster but much more expensive deployment option with US and EU data residency support.
Read news →Anthropic has released Claude Haiku 5.5 with lower small-model pricing, broad cloud availability, and a cache-read price cut for Sonnet 5.5, sharpening routing decisions for high-volume API workloads.
Read news →Cloudflare’s Web Search API beta lets developers route Ceramic.ai, Exa, and Linkup search through AI Gateway, adding logging and credit-based billing to live web grounding for agents and LLM apps.
Read news →Mistral Large 4 is now in public preview through Mistral’s API, giving teams a way to test the company’s new multimodal MoE model before planned open-weight availability later in October.
Read news →OpenRouter’s new Security Center gives organization admins a centralized view of risky API keys, including keys without spend limits, keys without expiration dates, and keys that have been idle for 90 days or more.
Read news →Amazon Bedrock Managed Agents powered by OpenAI is now in preview, giving AWS customers a way to run OpenAI-optimized agent workflows with IAM roles, CloudTrail logging, MCP tools, durable sessions, and human approval controls inside AWS.
Read news →OpenAI’s GPT-6.1 Sol is now available through the API with lower-cost positioning, tool calling, a new cached-input price profile, and beta multi-agent support, giving teams another routing option below GPT-6 Astra.
Read news →Cloudflare’s Auto Router public beta turns model choice into an AI Gateway function, using task classification, cost, latency, quality, and availability signals to select and fall back between eligible models.
Read news →