AI Card Bao GAME.AICARDBAO.COM

Industry News

OpenCode discounts GLM-5.3 and DeepSeek V4.1 Flash at a 40% discount: the latter only costs $0.12 per million tokens input

On the model price page of OpenCode, there is a "60% off" mark next to the GLM-5.3 and DeepSeek V4.1 Flash models, which means they are charged at a 40% discount from the original price.

The price after discount is as follows, the units are per million tokens:

GLM-5.3: Input $0.56, output $1.76, cache read $0.104.

DeepSeek V4.1 Flash: Input $0.12, Output $0.48, Cache Read $0.0024.

According to the four-fold discount, the original price of GLM-5.3 is US$1.4 for input and US$4.4 for output, and V4.1 Flash is US$0.3 and US$1.2.

OpenCode’s reference for this price reduction is Fireworks and Baseten, saying that its pricing is 60% lower than these two companies. It also highlights two points: that these models all run on their own GPUs, and that the processing throughput is higher than the other. The former point is related to whose hands the data passed through. The latter point does not give specific figures and can only be regarded as its statement.

The most interesting thing to look at is the cache read price of V4.1 Flash, which is $0.0024, which is one-fiftieth of the input price. The typical usage of programming agents is to resend the entire context in each round. Most tokens hit the cache, so the real bill is closer to the cache price than the input price. At this price, a long session with a high hit rate is basically just paying for output.

These two models come from Zhipu and DeepSeek respectively. They are both open weights and can be deployed by anyone. Hosts can only compete for price and speed. OpenCode used to mainly make programming agents in terminals. Now it buys its own cards to do inference, and then uses 40% off to attract people. The number of ways is similar to that in the early years when cloud manufacturers subsidized users to compete for users.

There is no written date on the page when the discount will end. If you plan to use it for a long time, it is best to make a cost estimate based on the original price.

Related