Hacker News Reader

Comment by jrflo | Hacker News Reader

Does anyone actually pay API costs out of their own pocket? It's about 10x cheaper to just get a codex or chat gpt subscription, it's so heavily subsidized compared to the API that I'm sure it would be cheaper to use frontier models on a subscription plan rather than paying API prices for deepseek flash.

View on HN

I do. For local dev work, I'm mostly using jetbrains' Junie, I can swap between a collection of models from google, openai, anthrophic.

I've had more than a few people tell me "oh, it's so much cheaper to use a $20 claude account" or "i've never hit a limit ever using my openai". Inevitably.. I end up reading/hearing "oh, I need to give it another couple hours to start using it again"... I've never hit that with my approach, even if it's costing me a bit more. Being able to work when I want when I have time has some value.

I also have openai and anthropic direct API billing set up for hosted and client projects that need to call out to an LLM service.

I thought Air was JB's multi-model interface? What is Junie? (I see the buttons, but am very confused by JB's AI offerings in general)

Is it worth the ~10x extra cost over the subscriptions? (This is obviously a leading question). Also, I think you can use OpenAI's subcription login with Air, but not Claude's.

It IS confusing. I'm not even using Air.

Junie is their agent(?) that can connect to multiple providers. AFAIK you can bring-your-own-key or buy a subscription from them that lets you use your credits across multiple vendors (openai, anthropic, google) across a handful of models from each vendor. So yeah, you can use your openai credentials in junie vs paying for jetbrains junie credits.

Is it 'worth it'? I dunno. I've not used dedicated subscriptions. I adopted Junie when it first came out, and it's been 'good enough'. As I say, I've never hit the 'oh, i have to stop for several hours until they let me in again' wall because of just using the API approach.

As my monthly spend goes up, I might try something else out, but... gpt6-luna just came out. That is likely going to cut my spend already.

Why not use a codex or claude subscription? If you use the entire usage allotment on the $200 plan it's about $2,000 in equivalent API costs. Switching providers may be valuable but it's quite literally an order of magnitude cheaper.

That's... $200. That's a lot for me. And I'm tied in to one set of models. Using something else (like Junie) I can swap and try models from multiple vendors.

I would investigate if the $20/month plan is good enough for your use: The token-based plans usually work out to be much more expensive than the monthly ones. Or get both the Anthropic and Codex $20/month plans. I don't know if this makes sense compared to your current use case, but there is a very good chance it does given how much the per-token plans cost. The primary use case for them is enterprises that aren't allowed to use the monthly subscriptions.

I only used that number because that's what I use. If you get the $20 codex plan that's equivalent to $100 in API costs, still a much better deal.

Is it equivalent if I'm time bound and have to stop working for X hours?

I do. Sharing training data with OpenAI gives me a lot of complementary tokens. I go above that but it's still quite economical and I pick the right model for the task (Luna for most).

>it's costing me a bit more

Would you mind sharing how much it's costing you? I've been wanting to use APIs from within Intellij, but I hesitate because of uncertainty about the cost.

Junie is $30/month, and I get $35 in credits. There are times I go over the limit, and you can 'top up' in increments of $5 (min). On a busy month (for me) I may have spent and extra $40. Initially I didn't hit the limit. As it's more of a daily driver, I'm averaging an extra $20, so... maybe $50/month. Give it a shot - the CLI version made it more usable.

20$ subscriptions offer “top up” too I’m pretty sure.

I do, 3-4$ a month of deep seek is enough for my usage

Same thing. I'm sure a subscription would be better value if I had enough usage, but I don't. With Deepseek or similar a few $ is quite a bit of usage already.

I load OpenRouter up and use models like GLM Flash 5.3, DeepSeek Flash 4.1, Luna, etc. And I often have random niche needs where I need a handful of calls for say, a really good image reader like Gemini Flash 3.8 or whatever. You can do a lot with $25 on openrouter or direct to chinese providers. I am cautious about what data I send overseas, but also like... just because it is in China does not inherently mean it is any less secure than a US provider.

I can't remember the last time any real recourse has mattered for companies getting breached or mishandling my data. Their stock just goes up and the govt just shrugs.

I pay $100 for Codex and it last about a day in the weekly limit - mostly Astra and Sol.

Then i got $20 into DeepSeek and i've been using those $20 for two weeks every day now. Use case is automating computer/browser use - Astra is really good at it, but very expensive, Sol and Luna haven't been that great at it, Deepseek as at about 80% of Astra but lasts forever.

Curious to hear more details about your harness and setup if you’re open to sharing.

I'm using the Deepseek harness with the deepseek computer use and deepseek browser integration, it's somewhat janky but it gets the job done.

Also for coding i use OMP directly with Deepseek.

It's been rather awesome because even maxed out $20 gets you very far in between Codex resets.

I also tried developing a small game where I have Astra use as many Deepseek workers and it needs and it only does the 3D objects and high level architecture and integration, but it goes way too quickly to see good results yet.

I do, but via OpenRouter. Outside of work my use cases are small and cheaper models do great job at those. I noticed even if I "burn tokens like crazy" I still pay less than any subscription available (a few $ a month).

But I guess if I had an agent vibecoding on it's own, I'd go with subscription instantly.

I use local LLMs on my Mac Mini.

Otherwise DeepSeek Flash 4.1 is dirt cheap (other "Flash" models are not that expensive either). I pay (very few dollars) out of my own pocket.

There are many things where having an API Key is necessary.

Maybe I’ve missed the boat though: is there now a method to use an api key to access a subscription?

You can't use an API key on subscriptions, but I've gotten around it using the `codex exec` command to run requests outside the CLI or GUI if you're already authenticated on that machine. Won't work for all cases, but I've never ran into a limitation in my use case of not having an API key.

I’ll look into this. But there is no universal solution. I find this to be quite limiting. I understand this is on purpose. If you want to automate, you need to pay more to have an api key.

It’s a good thing we have access to those very cheap and useful Chinese models.

> I use local LLMs on my Mac Mini.

which ones do you use?

I use the following models:

  - Qwen3.5 9B
  - Qwen3.6 35B A3B
  - Qwen3.8 27B
  - Gemma 4 26B A4B
  - Gemma 31B
  - Muse Glimmer 30B

I have 48G.

The MacMini is used solely for inference. llama.cpp + llama-swap.

People are saying local models are crap and serves no purposes.

I use them to help me spell check, write emails, write text messages, write JIRA tickets, write PR comments, etc.

I also use them in coding agents to complete different tasks.

I find them quite useful!

Thanks! I have two mac minis: one with 8 gb and the other with 24 gb (left with 17 gb useable ram); I can't run any open model on them. So, ordered a Mac 5 pro with 128 GB. Can't wait for its arrival.

128G of RAM? I envy you.

You will be able to run some pretty impressive models locally. Good for you.

I'm a small fish.

Some people here have 640 GB of RAM https://news.ycombinator.com/item?id=49621754

Large enterprises pay per token even through a ChatGPT “membership”, for example.

Which is why I said "out of their own pocket"

>heavily subsidized

Do you think this will be a problem in the future?

Reply on news.ycombinator.com