Anthropic’s latest flagship arrives, Xiaomi MiMo joins the catalog, every guardrail evaluation gets a full audit log, and deprecation dates are now visible across all providers.
Claude Opus 4.8
Anthropic’s latest flagship is now available via both the standard Anthropic endpoint and AWS Bedrock. Opus 4.8 supports a 1M token context window, vision, tool use, streaming, structured output, and reasoning effort levels (low / medium / high / max) — at the same per-token price as Opus 4.7, no markup.
Xiaomi MiMo
Two new models from Xiaomi’s MiMo family are in the catalog: mimo-v2.5-pro and mimo-v2.5-flash. Both support streaming, temperature control, and reasoning parameters.
Guardrails event log
Every guardrail evaluation now generates a logged event. The new Guardrails Events page shows a paginated, filterable view — filter by rule, direction (input/output), and decision (allowed/blocked). Each row shows evaluation timestamp, latency, what was matched, and the final decision.
Multi-region routing for Bedrock and Vertex
Pin any request to a specific geographic region using the X-LLMAPI-Region header — now supported for AWS Bedrock and Google Vertex in addition to Azure. Useful for data residency requirements or reducing latency for users in a specific region.