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.
Fireworks AI has introduced a half-price background inference mode, an ultra-fast tier, longer prompt caching and new governance tools. The releases make workload urgency, auditability and deployment location more explicit parts of AI routing and billing.
Read news →OpenAI now documents a way for automated workloads to exchange external cloud and platform identities for short-lived access tokens, reducing dependence on stored API keys while adding new identity-policy work for engineering teams, gateways and resellers.
Read news →Snowflake has moved Cortex AI Gateway inference, observability and cost management to general availability, giving enterprises a governed model endpoint inside their data platform and raising the bar for independent multi-model gateways.
Read news →GitHub is making Copilot available across GitHub Enterprise licenses through a usage-funded arrangement. The offer removes a separate seat-access decision, but most AI features still depend on an enterprise budget from the first request.
Read news →GitHub has introduced server-to-server authentication for Copilot SDK applications, allowing GitHub Apps to use short-lived installation tokens while attributing usage and costs to the organization. The preview removes the need to borrow a developer’s identity, although per-app budget controls remain limited.
Read news →Anthropic has introduced a beta organization-level API for retrieving Claude usage and cost data by model, workspace, API key and actor. The new interface could improve chargeback and reconciliation for gateways and resellers, although its beta status and migration requirements demand caution.
Read news →GitHub Copilot Fusion is now generally available, bringing model selection, cascades and critique workflows into the IDE. Its per-model billing and local-cloud preview create practical questions about cost controls, routing transparency and enterprise policy.
Read news →OpenAI has released GPT-6.1 Terra through its Responses and Chat Completions APIs at $1 per million input tokens and $5 per million output tokens. The model gives gateway operators a new price-performance option, but cached-input accounting, computer-use costs and a UAE residency restriction require more than a simple catalog update.
Read news →Google has launched Gemini 4 Argon in the Gemini API with introductory token pricing through November 30. For API gateways and enterprise teams, the release is not just a new model row: it changes catalog management, cache-aware billing, routing policy, and evaluation work across Google, GitHub Copilot, and Vercel AI Gateway surfaces.
Read news →Google has deprecated Gemini 3.7 Flash and Gemini 3.5 Flash API slugs and is automatically routing requests to newer replacements. For API gateways and developer teams, the change is less about a model catalog update than about whether logs, invoices and evaluations can show what was requested versus what actually served the request.
Read news →Cloudflare AI Gateway has standardized how it reports rejected provider credentials, changing some failures that previously looked like upstream server errors into explicit 401 responses. The update gives gateway operators, BYOK products and reseller dashboards a cleaner signal for invalid keys versus provider or billing outages.
Read news →OpenAI has released a public beta Decisions API for typed, low-latency classification-style calls using GPT-6 Luna. The new endpoint is not just another model listing: it gives gateways, analytics systems and reseller platforms a separate workload type to route, meter and document.
Read news →