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.
OpenAI has introduced the Agents API in public beta, giving developers a hosted way to run multi-step agent sessions with tools, MCP support, context compaction and parallel subagents. The launch changes what API gateways, observability systems and enterprise controls need to track.
Read news →Cloudflare AI Gateway has changed monthly usage invoices to show a single total-cost line per model and standardized provider/model identifiers across invoices and logs. The update simplifies reconciliation, but it also removes invoice-level input/output token granularity unless teams preserve it themselves.
Read news →OpenAI has begun rolling out GPT-6 Astra with a new API model ID, a million-token context window, published pricing and migration limits that remove several familiar sampling controls. For gateways and enterprise API teams, the launch is as much an integration event as a model release.
Read news →Kong AI Gateway 2.0 is now generally available with MCP server bundling, identity-aware AI policies and modality-aware cost controls. The release shows how AI gateways are moving beyond model proxying into governed access for models, tools and agent infrastructure.
Read news →Anthropic has released Claude Fable 5.1 with a new API model ID, broad cloud availability and much cheaper cache reads. For AI gateways and enterprise teams, the launch is not just another model row: it changes how repeated context, Batch API use and restricted access tiers should be priced and governed.
Read news →AWS has added fine-grained access control to Amazon Bedrock AgentCore Memory, using OAuth JWTs and Cedar policies to isolate agent memory by user or tenant. The change matters because agent state is becoming shared infrastructure, not just application data.
Read news →OpenAI says it plans to wind down Cursor’s direct access to OpenAI models after Cursor’s acquisition by SpaceX, with a proposed shutoff date of November 12, 2026. The move exposes a practical risk for coding teams: model access can disappear because of provider contracts, not only because an API or model is deprecated.
Read news →OpenAI has opened submissions for its WebMCP Challenge, promoting an experimental standard that lets websites expose structured tools to agents in ChatGPT’s desktop browser. If the approach takes hold, agent integrations may depend less on brittle screen automation and more on site-level tool schemas, permissions and audit trails.
Read news →Google Cloud has added more flexible Gemini Enterprise billing and new AI spend controls for agent workloads, including project-level pooled quotas and an AI Cost Summary Agent that can analyze Gemini API and Vertex AI spend by API key. The move shows that AI cost visibility is becoming part of the agent platform itself, not just a finance-team afterthought.
Read news →OpenAI has cut GPT-5.6 Sol API pricing by more than 20% for three months, lowering listed prices to $4 per million input tokens and $20 per million output tokens. The temporary change affects routing, reseller margins, cached-token accounting and how gateways pass provider discounts through to customers.
Read news →AWS has published Agentic Resource Discovery, an open specification for helping agents find tools, skills, MCP servers, APIs and other resources across environments. The move pushes agent infrastructure beyond routing and into catalog governance.
Read news →OpenAI’s deprecated Assistants API reaches its August 26, 2026 shutdown date, pushing remaining users toward the Responses API and forcing gateways, SDK wrappers and agent products to handle legacy endpoint failures clearly.
Read news →