From 53e0f024e57b698aefd1eccec93b8fcfe34bec7e Mon Sep 17 00:00:00 2001 From: dawn <90008@klbr.net> Date: Wed, 14 Oct 2026 21:54:08 +0300 Subject: [PATCH] cache-ttl: price argon cache reads at google's documented 95% discount Signed-off-by: dawn <90008@klbr.net> --- priv/cache-ttl.json | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/priv/cache-ttl.json b/priv/cache-ttl.json index 297e0531..eaf85db9 100644 --- a/priv/cache-ttl.json +++ b/priv/cache-ttl.json @@ -173,10 +173,10 @@ "write": 1.0 } ], - "read": 0.1, + "read": 0.05, "evidence": "measured", "source": "provider_requests: 625 gemini-4-argon calls, hit rate by gap after a call that hit", - "note": "Cloud Code Assist argon: after a hit the next call hit every time up to 7 min (535 of 539 under 30s, 35 of 35 from 1 to 7 min), missed the only two past 7 min. Few samples past 2 min, so the tier stays short of what was seen. The quota fit puts a cached token at 0.03-0.1 of an uncached one; 0.1 is the documented discount. Each variant id caches apart.", + "note": "Cloud Code Assist argon: after a hit the next call hit every time up to 7 min (535 of 539 under 30s, 35 of 35 from 1 to 7 min), missed the only two past 7 min. Few samples past 2 min, so the tier stays short of what was seen. Google documents a 95% discount on argon cache hits, and the quota fit puts a cached token at 0.03-0.1 of an uncached one. Each variant id caches apart.", "checked": "2026-10-14" }, { -- knotmirror 0.0.1