Every release that touches an externally-visible surface is logged here. Internal-only changes (replica rebalances, observability plumbing) live in our SRE log.
The 1M-context Gemini frontier model is now generally available across all routed regions. Streaming, tools, vision, and structured JSON all supported on day one.
gemini-3.1-pro in the model catalog
Set idc-no-retention: true on any request to force a route where
prompt and completion bytes never leave volatile memory. Available across all
frontier and balanced-tier models.
retention_class fieldThe new Anthropic frontier model is live across all routed regions. Claude Sonnet 4.6 (shipped February) remains our recommended balanced default for agentic loops thanks to its tool-use stability across long multi-turn sessions.
claude-opus-4.7Sixth routed region. Australian and New Zealand customers now route locally instead of via Singapore — typical p50 latency improvement is 60–90 ms.
ap-syd region hintScale and Enterprise plans can now define rate budgets per downstream tenant. Useful for multi-tenant SaaS products that want to prevent one heavy account from starving the rest.
POST /v1/management/tenantsStream per-request traces directly to your existing OTLP/gRPC collector. No proprietary dashboard, no parallel observability stack.
us-east-1, now us-east)Two more open-weights flagships join the efficient tier. Both are available on shared capacity and as dedicated replicas in EU and APAC.
deepseek-v3.2qwen3-maxBring your own upstream key under our schema, routing, and observability layer. Keep your negotiated rates with the provider; pay us only the per-request gateway fee.