Analysts dispute claim that Chinese LLM providers are ending open source as they raise API prices

> raising prices or gradually shifting toward more closed models entirely separate things. A, token dumping was always just a response to DeepSeek. B, prices are driven by scarce compute, this makes them *more* likely to open source. C, Alibaba != scare-quote open source players

Well, scarce compute and ROI incentive of course. But the I part is ballooning because of how scarce the compute is. ZAI literally can't serve all their subscription demand even at "high" prices, so they lose relatively little by open sourcing.

@zephyr_z9 but how long they will be able to serve models at dirt cheap price? as now they also have raised funding so given that they are now have answer those investors as well.

@zephyr_z9 and they have already increased price for the upcoming July launch.

@zephyr_z9 They too doubled the pricea during peak hours : DeepSeek adds peak-hour surge pricing to its AI API https://thenextweb.com/news/deepseek-peak-hour-api-surcharge-v4-price-war

@heyiamsenku @zephyr_z9 I dont think the price increase was a move to increase revenue. They are using price as a means to move usage to other periods of the day, thus increased pricing in peak (6hrs) to move usage to off-peak times.

@teortaxesTex Isn’t that a bit too idealistic as a counterargument?

If GPU scarcity is pushing them toward open source, doesn’t that also mean GPUs for training are scarce as well?

And if that’s the case, then the cost of building LLMs themselves is going up.

@zephyr_z9 Let's see who's left when the hype fades. 🐋

@KailashNar50501 @zephyr_z9 Clearly an effort to move usage to the other 18hrs of the day where the compute is under utilized.

@heyiamsenku @zephyr_z9 Imo we will see further efforts by deepseek to improve efficiency and reduce pricing. They so no point in intelligence being expensive that only a few can access it.