●ROBOTICS — Gemini Robotics ER 2 is in public preview, covering spatial reasoning, agentic code execution, and multi-robot coordination●STREAMING — gemini-robotics-er-2-streaming-preview targets real-time streaming over the Live API, with bidirectional audio and video input●SUNSET — Shutdown dates are close: Imagen models on August 17 (gemini-3.1-flash-image is the successor), the Grok 4.1 family on the 20th, and gemini-robotics-er-1.6-preview on the 31st●SAMPLING — temperature, top_p, and top_k are now deprecated. If your app exposes them as user settings, that surface needs rethinking●LOGS — The Interactions API now supports developer logs, viewable from the AI Studio dashboard●MODELS — Gemini 3.1 Pro remains in preview. 3.6 Flash reached GA on July 21, and 3.5 Flash-Lite suits high-volume subagent work●ROBOTICS — Gemini Robotics ER 2 is in public preview, covering spatial reasoning, agentic code execution, and multi-robot coordination●STREAMING — gemini-robotics-er-2-streaming-preview targets real-time streaming over the Live API, with bidirectional audio and video input●SUNSET — Shutdown dates are close: Imagen models on August 17 (gemini-3.1-flash-image is the successor), the Grok 4.1 family on the 20th, and gemini-robotics-er-1.6-preview on the 31st●SAMPLING — temperature, top_p, and top_k are now deprecated. If your app exposes them as user settings, that surface needs rethinking●LOGS — The Interactions API now supports developer logs, viewable from the AI Studio dashboard●MODELS — Gemini 3.1 Pro remains in preview. 3.6 Flash reached GA on July 21, and 3.5 Flash-Lite suits high-volume subagent work

TAG

p95

1 articles

← Back to all tags

Gemini API¹ Tail Latency¹ SLO¹ Streaming¹

◈ Gemini API/2026-06-23Advanced

Your Gemini API Average Latency Looks Great — But Some Users Still Get Stuck. Defending p95/p99

Your average TTFT is fast, yet a fraction of users keep hitting frozen responses. That is a tail-latency problem (p95/p99). From measurement to model routing, streaming budgets, cache accounting, and retry design — here are the defenses that actually held up in production, with code.