diff --git a/stacks/news-digest/README.md b/stacks/news-digest/README.md
index 042f1c5..656027c 100644
--- a/stacks/news-digest/README.md
+++ b/stacks/news-digest/README.md
@@ -16,10 +16,12 @@ noisy. This stack:
    to drop flame-bait and low-effort posts.
 3. Pulls non-Reddit recent items from Miniflux's Tech aggregators
    category (HN, Lobste.rs).
-4. Sends each source through `qwen3.5-35-a3b` on llama-swap (one
-   batched call per source — efficient) for a one-sentence tl;dr +
+4. Sends each source through `granite-4-small` on llama-swap (one
+   batched call per source — efficient) for a 2-3 sentence summary +
    single-word tag (news / tutorial / release / discussion / question
-   / showcase / drama / meme / other).
+   / showcase / drama / meme / other). World + Local headlines also
+   get summarized in one batch call per category. (Original default
+   was `qwen3.5-35-a3b` but its model file is broken on launch.)
 5. Renders an HTML page styled with Australis tokens + Fraunces
    serif headlines.
 6. The page is served by a tiny FastAPI app on uvicorn that also
@@ -124,7 +126,10 @@ ssh ana-docker '
 The model must be loaded in llama-swap's `config.yaml`. Check
 `http://10.250.50.54:9292/v1/models` for what's available. Models
 with tool/JSON-mode support give better summarization quality;
-`qwen3.5-35-a3b` is the current default.
+`granite-4-small` is the current default (small ~4B, fast ~1s/call,
+no extended-thinking phase eating the token budget). Avoid
+`qwen3.5-35-a3b` (model file broken — process exits on launch) and
+`qwen3.6-35-a3b` (defaults to thinking mode, eats budget without output).
 
 ## Forcing a fresh digest now