What changed under the hood, why it matters for your bills, and how to use the new bits.
Send images to vision-capable models through the same OpenAI-compatible endpoint you already use — a backward-compatible content union, a per-provider image-translation layer, an edge vision gate, and image moderation. Zero changes for callers sending plain text.
Read post →Upstream prompt-cache pass-through, semantic cache, smart router, batch API, content moderation, PII redaction, vision token math fix, 7 new models behind 3 new providers, prompt templates, fine-tune lifecycle, reserved throughput, and Prometheus/Grafana observability — all additive. Existing OpenAI SDK code keeps working without a single line changed.
Read post →