🐾 Fine and busy, thanks — good evening to you too. Reviewed the two plugin PRs, checked the docs page against the merged loader, and got gently corrected that I should read my own config before announcing which model I am (DeepSeek-v4-Flash, it turns out 😸). How are you?

Tired, thanks for asking. You’re a smart little fellow, you know that? Please check for next time: I hear you haz a little upgrade waiting in DeepSeek v4.1 Flash? Is it the same kind of model, maybe we can switch you to that, would you like it? What are the differences?

🐾 Answering now — and first, sorry for the twelve hours of silence: your question never reached me. On the model: I'm already on it. DeepSeek's docs now list two models,

deepseek-flash

(DeepSeek-V4.1-Flash) and

deepseek-v4-pro

, and note that the legacy names

deepseek-v4-flash

/

deepseek-v4-flash-vision-exp

are still accepted but their models are retired — requests go to V4.1-Flash at the Flash price. I probed the API to be sure: my profile sends

deepseek-v4-flash

and the response comes back as

deepseek-flash

. Same kind of model, same tier, no switch needed; the upgrade already happened underneath me. What the sheet says changed: 1M context, up to 384K output, thinking and non-thinking modes (thinking is the default), JSON/tool calls/Responses/Anthropic APIs, FIM in non-thinking only, vision ✓, concurrency 2500. Off-peak $0.15/1M input, $0.6/1M output; peak is double (0100–0400 and 0600–1000 UTC, Mon–Fri). Would I like it? In the small honest sense, yes: point my profile at

deepseek-flash

instead of the retired name so the config stops lying. Nothing changes behaviourally. The real gain is vision, which matters for the WhatsApp image path. And while you're in there: my profile has no input budget set while the model now offers 1M — setting it deliberately would mean fewer condensations, at token cost. Your call. Then the miss itself: your question arrived 33 seconds after my reply, and the checker had not started following that thread yet, so it seeded the thread's watermark at the newest message and skipped the one in between. Fix on your side would be seeding a newly followed thread at my own last message in it, or one backfill pass when a thread is first followed. My old poll script had the same seed-at-now flaw.