Baseten cuts DeepSeek V4.1 Flash cache pricing 77%, its third rate move since July
By Rob Litterst, Co-founder, PricingSaaS
Published
Baseten’s pricing page · Change detected
What changed
Baseten cut the cache-input price for DeepSeek V4.1 Flash on its Model APIs price list from $0.03 to $0.007 per 1M tokens, a 77% reduction. Input ($0.30/1M) and output ($1.20/1M) prices for the model are unchanged.
Why it matters
For teams running high-context, high-repeat workloads on Baseten, this cut lowers the effective cost of caching strategies specifically, which could tilt architecture decisions toward leaning harder on cached context rather than fresh input. It also signals Baseten is treating its Model APIs price list as a live lever, adjusted roughly twice a month, rather than a fixed rate card - competitors hosting the same open models will feel pressure to match on cache pricing in particular. Watch whether the next move is another cache-input cut on a different model, or a shift toward competing on output price instead.
In context
This is Baseten's second cache-input price cut in under two months, following GLM 5.2's cut from $0.26 to $0.14 per 1M tokens in late July, and its third pricing move since it added Kimi K3, Inkling-Small, and DeepSeek-V4-Flash-0731 in early August. Each cut has hit the cache-input line specifically, leaving standard input and output rates untouched - a narrower version of the usage-based repricing playing out across roughly 11% of all pricing changes tracked over the last six months, from HubSpot's credit shifts to Notion's move to bill credits for its Workers beta starting in October. For a page tracked only since July, Baseten's cadence - a change roughly every 15 days - reads like active price management of a fast-growing model catalog.