GEMINI LABJP
IMGEOL — The older image generation models are deprecated and shut down on August 17. Move to the newer stable or preview endpoints before thenREGION — Gemini 3.5 Flash disappears from the global region in the Gemini Enterprise app on August 4, so check any config that pins region and model togetherGROK — The Grok 4.1 model family on the Gemini Enterprise Agent Platform is deprecated and will be shut down on August 20FSMM — File Search now handles multimodal search through gemini-embedding-2, letting you index and query images natively instead of transcribing them firstFLASH36 — Gemini 3.6 Flash and 3.5 Flash-Lite are generally available. 3.6 Flash improves token efficiency and agentic planning while sitting below 3.5 Flash on priceSAMPLING — The temperature, top_p, and top_k sampling parameters are now deprecated, so any output control built on them needs a different approachIMGEOL — The older image generation models are deprecated and shut down on August 17. Move to the newer stable or preview endpoints before thenREGION — Gemini 3.5 Flash disappears from the global region in the Gemini Enterprise app on August 4, so check any config that pins region and model togetherGROK — The Grok 4.1 model family on the Gemini Enterprise Agent Platform is deprecated and will be shut down on August 20FSMM — File Search now handles multimodal search through gemini-embedding-2, letting you index and query images natively instead of transcribing them firstFLASH36 — Gemini 3.6 Flash and 3.5 Flash-Lite are generally available. 3.6 Flash improves token efficiency and agentic planning while sitting below 3.5 Flash on priceSAMPLING — The temperature, top_p, and top_k sampling parameters are now deprecated, so any output control built on them needs a different approach
TAG

regression testing

1 articles
Back to all tags
Related:
Gemini API1temperature1migration1cost design1
Gemini API/2026-08-01Advanced

When Sampling Parameters Were Deprecated, Diversity Broke Before Determinism Did

Deprecating temperature, top_p and top_k hurt the diversity-generating side of my pipeline, not the deterministic side. Counting real call sites, moving diversity to the input layer, and measuring effective diversity.