TL;DR — MiniMax M3.1 Flash is the new MiniMax default ($0.15/$0.60, 1M context), and DeepSeek's deepseek-flash is now V4.1-Flash-1017 with no change to price or id.
Added
- MiniMax M3.1 Flash (
MiniMax-M3.1-Flash) on all three MiniMax providers
(minimax, minimax-api, minimax-cn), first in the picker and the new default. 1M context; on the pay-per-use API it is $0.15 in / $0.60 out with cache reads at $0.015. MiniMax M3 stays in the picker, and a config that names it keeps it. - /thinking now works on it. Low, Medium, High and Max map to the five levels MiniMax documents (it also has xhigh, which no tier sends). Thinking cannot be switched off on this model, so no tier does. /thinking auto sends nothing and the model runs at its own default, which is max: the heaviest level, and the one that spends a subscription's 5-hour window fastest. Pick Low, Medium or High for everyday work. MiniMax M3 has no such control and gets none. The Mac app has the same control. - The retired MiniMax-M3.1-Flash-Preview name is served by this model.
Changed
- DeepSeek V4.1 Flash is now the 2026-10-17 release (V4.1-Flash-1017).
DeepSeek kept the id (deepseek-flash) and the price, so nothing changes in your config or your estimates; the picker description names the version. OpenRouter lists this snapshot as its own id, deepseek/deepseek-v4.1-flash-1017.
Fixed
- MiniMax M3 cache hits were estimated at half their price. A cached token
costs $0.06 against $0.30 for a miss (0.2×), and Codeep billed it at the 0.1× default. /cost and /stats for M3 sessions with a lot of caching were low by that difference. The Mac app had the same gap and is fixed alongside.