OpenCode Go plan which STILL advertises that it gives users $60/month of credit for each $10/month sub making it one of the best LLM plans out there, is no longer actually doing that. They are now giving users a single pool of $15 of credits for each $10/month sub for all their new models: GLM-5.3, Kimi K3, DeepSeek V4 Pro 0813, and DeepSeek V4 Flash 0731.
This is in spite of new DSV4 and GLM models actually remaining exactly the same as the previous versions but with post-training (same hardware and memory requirements), there's no practical reason for these models to increase prices from both OpenCode and DeepSeek other than sacrificing their goodwill and userbases to increase profit.
This especially applies to DeepSeek V4 Flash as its one of the least demanding on hardware. Some older models still offered by the opencode Go plan at $60 credits require significantly more hardware! (MiMo-V2.5 is a larger model for example)
And the worst part is that opencode still (til this moment) write in their Go FAQ that they offer $60/month credits so this is less than honest on their part:
4 Comments
onlinepersona@programming.dev · 1 pts · 23h
I do wonder if it wouldn't be better to pay an inference provider and then select the models you want. It's surprising to me that opencode had that offer at all. Where did they get the money to subsidize that?
davidagain@lemmy.world · 3 pts · 23h
Investors. Most LLM companies are making massive, incomprehensibly bad losses and are trying to train large customer bases to rely on their LLMs before it turns profitable. Prices are going to have to multiply by least ten.
https://isaiprofitable.com/
SirDimples@programming.dev · 1 pts · 21h
Chinese open weights models are no longer niche. I think opencode were betting on the niche angle and over-subscribing far above their capacity to serve hoping most users don't saturate their quotas. But this is no longer the case thanks especially to the new deepseek v4 flash which kinda broke the economy of API resellers.
Personally, I can understand the economic complexity combined with the huge surge in demand. What I don't understand is why be vague about serving capacity and pricing while continuously tweaking usage quotas significantly without any notification to their subscribers.
ICastFist@programming.dev · 2 pts · 19h
Since no AI company is publicly traded yet, they have no requirement to show their balances, which everyone suspects is deeply on the red. They can't even manage to secure enough venture capital to pull what other big tech companies did, of dominating a market first before exploiting the userbase for maximum profit, they'll go under much sooner than any of them can even smell a hint of profit