DeepSeek price hike in practice: cache-hit tokens up 6-12x, nightly batch bills quadruple
CryptographerThen999 · reddit · 2026-08-26
A small team shares the real invoice impact of DeepSeek's new pricing. Their nightly support-ticket summarization batch hits >90% cache, yet V4-Pro cache-hit pricing rose from 0.003625 to 0.022 off-peak and 0.044 peak (6-12x), and since peak windows map to Beijing office hours, their unmovable 06:30 UTC cron lands at double rate — total bills are now 4-5x for identical work.
The author notes the schedule is nearly free for California users (peak = US night) and is trialing MiniMax M3's fixed monthly token plan: cached reads draw 1/5 of a fresh token and rates don't vary by hour, though the pool doesn't roll over. Lesson: cheapest per-token pricing matters less than a rate you can budget on.
More from Models
- Unredacted Anthropic filing exposes internal data-quality debate months before Opus 3 cutoff — zacharynado · 2026-10-15
- Google says teams across the company are using Gemini 4 Argon on its hardest engineering problems — CurieuxExplorer · 2026-10-15
- Perplexity's new embedding models: index with a 9B model, search with 0.6B — antoine_chaffin · 2026-10-15
- 797GB of n-gram counts in CPU RAM let 1B models match 57% more compute — PangWeiKoh · 2026-10-15
- Leak: Moonshot's Kimi flagship 'x4' done with pre-training, K3.1 in internal testing — lyraxana · 2026-10-15
- Unsloth ships GGUF quants for Nemotron 3.5: Bolt-74B runs in 24GB RAM — danielhanchen · 2026-10-15