Requesty launched five new models in three weeks: GLM-5.3 Flash, Gemini 3.8 Flash, Claude Fable 5.1, and GPT-6 Astra.

Requesty is the control plane for AI spend and access. Teams point their existing OpenAI-compatible code at one endpoint and reach 600+ models from every major provider. No new SDK, no provider-by-provider integration work, no separate contract for every lab. From there the platform does what production demands. Requests fail over down a fallback chain when a provider degrades. Routing rules move traffic by cost, latency or capability. Caching and server-side prompt management keep repeat work off the bill and prompts out of the codebase. MCP support extends the same governed path to tool-calling agents. Model choice becomes configuration, so moving to a better model is a decision rather than a release. Every call is logged and costed, so engineering and finance read the same number: spend by model, by key, by user, by group, by application. Per-key, per-group and per-user budgets stop a runaway job before it reaches an invoice. As teams grow, the same layer carries the controls their own buyers ask about. Enterprise plans add SSO, role-based access control, audit logs, approved-model allowlists, team budgets, guardrails and PII detection. Teams with European requirements can route through a dedicated EU gateway with in-region processing and zero data retention by default. Companies do not adopt a gateway because calling one model is hard. They adopt one because the second model, the second team and the first invoice arrive at the same time.