MiniMax-M3.1-Flash API pricing and cost calculator

How this is calculated

The calculator opens on a request of 50,000 input tokens, 3,000 output tokens, half the input cached, and 500 requests a month. One uncached call costs $0.0093. With half the input read from cache it's $0.005925, and the starting total is $2.9625. Inside the off peak window the same half cached call costs $0.0029625, with input at $0.075, output at $0.30 and cache reads at $0.0075 per million tokens. That's 50% off input and 50% off output. The hours aren't typed here because MiniMax can change them. Read MiniMax's pricing overview for the current window, and convert it to the time zone your scheduler runs in. A nightly job is the obvious fit, so a job that sends 5,000 requests of 50,000 input tokens and 3,000 output tokens, half cached, costs $14.8125 off peak against $29.625 at the standard rates. Other shapes help too. An agent session of 30 turns with 80,000 input tokens each, about nine tenths read from cache, and 1,500 output tokens per turn costs $0.0954. A document intake step that reads 100,000 scanned pages as images is harder to price. Image and video inputs turn into tokens in ways the provider decides, so send a handful of real samples and read the usage counts that come back before you multiply. Don't scale an estimate from text only tests. An API bill is per token however the model handles attention internally, so a 700,000 token prompt costs $0.1062 uncached at the standard rates. The prices shown are the maker's own. The weights are open, so other hosts can list the model, and the calculator will show a lower host price here if one appears.

Verdict

Treat MiniMax-M3.1-Flash as a model with two price lists. Plan interactive traffic on the standard rates and move scheduled work to the off peak window only if the saving is worth the scheduling effort. Because the input side takes images and video, run a small sample of your real media first, since the token count per file decides the bill more than the rate does. The weights use the minimax-community license, so read it before commercial use. For self hosting, the VRAM calculator page for this model shows how much of the budget is cache at long context.