Rendered at 21:29:36 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
Frannky 6 hours ago [-]
I'm using oh my pi with Kimi K3 as the planner and DeepSeek Flash 0731 as the implementer, with OpenRouter as the provider.
Any suggestions for a better configuration? I mostly need Opus 4.6 + Claude Code alternative. I don't need Fable level capabilities, I add one new feature at a time and approve the code before shipping it. Then test and open a PR.
Usually both Fable and Opus suggest dumb ideas but are good implementers once I tweak the idea, approve the code, and add unit and production tests.
I'm OK spending max $100/month on APIs, ideally with Zero data retention. I only need a few hours a day of coding. I don't want agents running all the time; I figure I can stay on top to each feature and wrap my head around the product and new suggestions as long as I don't build too much at once.
I'm still on a Claude Max $100 plan, but it's barely usable anymore—one call and I hit 20–30% of the 5h window on Opus 4.8. Opus 5 seems tuned to make messes, and Fable burns tokens for a level of capability I don't actually need.
anentropic 6 hours ago [-]
I'm curious what kind of prompts people are doing to get 'dumb' ideas out of Fable
Frannky 5 hours ago [-]
It's usually one of two things. First, I say I want to build this system solving this problem with these requirements, what do you suggest? It suggests X, then I say, "Hmm, why not Y?" and it goes, "Oh yeah, Y is better," etc.
It's similar to the situation where you ask whether you should walk or take the car to go refill the car with gas. I feel LLMs have a linearity embedded in them that prevents them from finding non-linear, smart solutions, especially when data is scarce. They can probably get it done, but with far more complex solutions, which increases the risk of ending up with a crazy complex codebase when it could have been much simpler and more elegant.
The other situation is when I spot a problem or a feature change that's needed after experiencing the product, and I ask how to change the codebase, it suggests something, I say "why not this other thing," and then we do the other thing.
throwaw12 6 hours ago [-]
not related to your question directly, but noticed you mentioned oh-my-pi, can you share little bit more why you went with oh-my-pi and not install your own set of extensions? (asking because I was just looking at it to enhance my workflow, but feeling it has too many things)
alhimik45 2 hours ago [-]
I was in your situation some time ago. Good thing is that advanced (but little bloated, yes) harnesses have good enough defaults, so you don't have to go and read each setting before the use.
Regarding Pi my position is that it is brilliant piece of software if you don't need extensions - give the model bash tool and let it do all it wants through it or use Pi as SDK for your own advanced harness or smth similar.
But Pi with extensions has two problems. First one: rather often they don't play well together. For example, if you want some adjustments A and B for the same tool and there are two extensions which do A and B, they will likely not work as expected when installed simultaneously. You could say that it can be solved by adjusting extensions or just generating your own - yes and it is the second problem. Like any piece of code you own and use, you have to maintain it. Bug here, incompatibility there and voila - you spend your precious time to work on harness instead of doing your job. Plus remember that vibecoders are not very responsible people, so Pi extensions registry is flooded by "use Pi to customize Pi" buggy one shot extensions.
With carefully developed set of extensions Pi would be better, like properly configured Arch Linux could be better than Linux Mint in the hands of power user. But considering how fast things are changing in this sphere, seems it is more optimal to take more bloated harness - with unneeded tools, too big prompts, etc - which will be effective on 90%, but do the actual job with that harness right now.
DrammBA 6 hours ago [-]
It's all about trade-offs, do you want a bunch of extensions of varying quality from a bunch of developers, or do you want a bunch of extensions of varying quality from one developer?
actsasbuffoon 5 hours ago [-]
Or the third path: use Pi to customize Pi. It’s pretty fast and easy, and now it does exactly what you want.
DrammBA 2 hours ago [-]
That's just "a bunch of extensions of varying quality from one developer" with extra steps
Frannky 5 hours ago [-]
I just didn't want to spend time configuring an harness
yojo 6 hours ago [-]
Honestly, GPT 5.6 Luna is worth a look. It’s a reasonably good implementer at a small fraction of the cost. $100 buys a heck of a lot of it at API pricing.
Not sure about the OAi Pro plan, doesn’t look like the 80% Luna price slash made its way into the quota system.
You could also try tuning down the effort level on Opus. It makes a huge difference in token consumption and you might be able to get away with lower than you’ve set
If you prefer pay-as-you-go, then Xiaomi MiMo is the only other provider with comparable models (MiMo v2.5 & Pro) that matches DeepSeek's current API rates for input/output/cache: https://mimo.mi.com/docs/price/pay-as-you-go
I have the 20x. I really never reach the limit anymore. I am a heavy user. How is that possible?
rvba 6 hours ago [-]
Do you import old code base or work on small greenfield projects?
system2 6 hours ago [-]
I agree with you. I am a 5x user; compared to 6 months ago, I rarely reach session or weekly limits. It got slightly slower, but fewer mistakes make up for it. Probably the commenter above is a heavy vibe coder launching a bunch of background agents to reread the base code over and over.
copperx 4 hours ago [-]
I'm on 5x also, and lately (this week), the window seems more generous, even on Fable. I wonder what's going on.
refulgentis 6 hours ago [-]
Are you using Kimi w/expectation of ZDR? Their EULA is “we use your data and train on it unless you negotiate something with us privately”
ncallaway 6 hours ago [-]
Fireworks.ai has ZDR and hosts Kimi K3
danvdb 6 hours ago [-]
[dead]
petercooper 7 hours ago [-]
I'm guessing this is almost entirely about the incredibly low cache read prices. Few have come close to them, nothing has a bigger effect on (a typical) session price, and with the price of RAM right now, they have to be the biggest pain point for them right now? A 10x increase in cache read would be a significant increase, yet would still keep them cheaper than every other provider of their model (at least based on the prices at https://openrouter.ai/deepseek/deepseek-v4-pro#providers)
cedws 6 hours ago [-]
I guess the low pricing was so DeepSeek could gather training material. The provider on OpenRouter indicates that data can be retained.
npn 6 hours ago [-]
From the private conversation of deepseek CEO and the investors, I'm under impression that they take pride for the low API price though. So it is not only training data. If they really want the data there are many other ways to do so.
andai 6 hours ago [-]
I wanted to say luna's way cheaper now too, but oh my god I must have misremembered, DeepSeek cache read is 0.002, not 0.02.
jaggs 5 hours ago [-]
Yes exactly this. Even a significantly higher price will mean they're way cheaper than rivals. I wouldn't make any changes until we see exactly what the price rises are.
efficax 49 minutes ago [-]
Not surprised, DeepSeek v4-flash is basically free right now. I've been using it for a custom agent that is more about orchestrating a lot of the "chores" of using a computer (calendar checking, reading slack channels for me, reading my email etc) and it's basically free. I think I've spent $2 in API credits in the past 60 days
While it won’t cover everything DeepSeek does, it handles sophisticated tasks quite well. I found it after spotting it on a chart of different models and it was listed as being near Deepseek v4 Flash’s price/performance levels.
I have noticed by the way that DeepSeek’s API has been pretty slow the past couple of days. This feels like a demand-driven move more than anything. Good thing I invested in an eGPU!
ignoramous 4 hours ago [-]
> Good thing I invested in an eGPU!
Curious: Which one?
> MiMo-V2.5-Pro to be a pretty cost-effective alternative ... I found it after spotting it on a chart of different models and it was listed as being near DeepSeek v4 Flash's price/performance levels.
MiMo v2.5 Pro is at DeepSeek v4 Pro price level (but consumes lesser tokens per task, so cheaper overall). DeepSeek v4 Flash costs ~3x lesser than the Pro variant!
shortformblog 2 hours ago [-]
I got the Gigabyte 5060ti AI Box, which is good but a real pain to set up on Linux. Not enough to run Deepseek but a lot of good options.
If you log and view your "Usage" page there is a blue banner near the top in a blue box saying:
> We plan to raise the overall pricing for DeepSeek API services in the near future, with a significant increase expected. Please plan your usage accordingly. The specific pricing plan will be subject to official notice.
patates 7 hours ago [-]
Well it was nice while it lasted. Built so much stuff for basically free. I can only thank them for that.
I guess this means that this powerful model for very cheap concept does not work?
mosura 6 hours ago [-]
They were showing notices about doubling pricing during american working timezones a week ago, so it seems likely to be demand related.
Their v4 flash is brilliant, and I expect they have had a huge, and expensive to handle at short notice, surge in customers as a result.
trollbridge 6 hours ago [-]
You mean Beijing working time zones?
mosura 6 hours ago [-]
No?
trollbridge 2 hours ago [-]
The one I saw was 9a-12n and 2p-6p China time. (China is one time zone.)
Nearly irrelevant for US/Eastern unless you’ve got major insomnia.
f6v 6 hours ago [-]
> I guess this means that this powerful model for very cheap concept does not work?
Remember how cheap Uber rides were? That being said, current models are absolutely incredible. And I’ll think they’re going to get cheaper when the next generation is released.
nerdix 6 hours ago [-]
It was insane how much you could get done for a couple of dollars.
The good times never last.
anigbrowl 6 hours ago [-]
Good grief, you guys make it sound like like they're going up 100x. I'm pretty sure it will continue to deliver good value.
anigbrowl 6 hours ago [-]
I don't think you should submit links to logged-in dashboards on HN.
For those who don't have a Deepseek account, this is just a banner reading:
We plan to raise the overall pricing for DeepSeek API services in the near future, with a significant increase expected. Please plan your usage accordingly. The specific pricing plan will be subject to official notice.
storus 6 hours ago [-]
2x DGX Spark runs DeepSeek V4 Flash 0731 quite well and I already reimplemented a few classical games with it in a few hours in OpenCode with minimal hand-holding. I am expecting the inference price to collapse over time and not increase, at least for models that are able to do like 95-99% of work quickly and only the most problematic parts requiring something better.
tempoponet 6 hours ago [-]
It's exciting, but that's $9k for a model that's in the ballpark of last year's frontier. For many people that's like 10 years of up-front cost in a landscape that's getting better and cheaper every week.
but compared to the $500k+ racks running the current frontier, it does give perspective.
vitaflo 7 hours ago [-]
I have to assume this means that they are about to release the final version of v4 Pro and it uses more resources than the preview did
recov 7 hours ago [-]
Bound to happen. I’ve been using the new flash and it’s insane the value I’ve gotten, I wish I used it more for some other homelab things.
Readerium 3 hours ago [-]
I expect 5X cache input price hike, and 2X usual input/output hike.
midnightbobarun 6 hours ago [-]
DeepSeek was such a workhorse... I've had some pretty decent results with Hy3 and Nemotron too, but they're not the same. I'll probably have to switch to them (or find other alternatives) depending on how much DeepSeek raises prices.
GTP 6 hours ago [-]
The link takes me to a login page.
timpera 6 hours ago [-]
> We plan to raise the overall pricing for DeepSeek API services in the near future, with a significant increase expected. Please plan your usage accordingly. The specific pricing plan will be subject to official notice.
It would be nice to have this notice on the public documentation as well.
elmer2 6 hours ago [-]
Even if DeepSeek is better than the American models, I won't be using it unless there is a flat-rate monthly plan. This is only real way it's useful.
If not, the cost outweighs whatever value I might have gotten from it.
surgical_fire 5 hours ago [-]
lol it doesn't.
I put 10 bucks on DeepSeek almost 3 months ago.
I still have 2 bucks there. I think I used so far something in the vicinity of 300M tokens total.
Even if they double their price it is still cheaper than US models on their flat rate plans, and I don't get locked out for expiring the quota.
abdullahkhalids 6 hours ago [-]
Is there currently a lot of difference between deepseek's own prices and other provider's prices of deepseek models?
Yeah, everyone else's cache read prices are 10x higher.
LoganDark 7 hours ago [-]
I guess they invested in some new infra and want to make that back over the next 10 months [0]:
> For us, a reasonable profit means roughly this: we buy a batch of servers, and we recover the cost in about ten months. Given the risks and the upfront investment, even if we depreciate a server financially over three or five years, commercially we think a ten-month payback is enough. That is the logic behind our current API pricing. For V3.2 Flash and other models, the standard is the same: recover the cost of the equipment in ten months.
Looks like a second order effect to them not raising their round
miroljub 7 hours ago [-]
From the announcement:
We plan to raise the overall pricing for DeepSeek API services in the near future, with a significant increase expected. Please plan your usage accordingly. The specific pricing plan will be subject to official notice.
Now, the question remains, what does that "significant" mean? Would it still be cheaper than the competition, or did they realize they are too cheap for what they offer?
Given that many inference providers offer DS4 flash for more or less the same input and output token price as DeepSeek, they have good profits even with todays low prices.
_aavaa_ 7 hours ago [-]
It’s all about the cache prices. Those dominate usage, cached tokens are easily >90% if not >95% of all used tokens.
miroljub 7 hours ago [-]
Yes, but I'm not sure. They recently dropped cache prices 10 times. Why would they rudder back after only a few months?
_aavaa_ 6 hours ago [-]
Don’t know but that’s not my point. My point is that even small absolute value changes in cached token prices will have massive impacts on final cost because of the sensitivity on it. If v4 flash cache prices go from 0.0028 to 0.01, the absolute change looks not that big (and would be a rounding error for input and output costs) but it would explode actual usage costs.
faangguyindia 6 hours ago [-]
to fund training a far better and capable model.
HarHarVeryFunny 6 hours ago [-]
Doesn't make sense - they very recently said they have plenty of money, and are only limited in training a larger model by lack of GPUs (Huawei as well as NVIDIA) available to buy.
jLaForest 7 hours ago [-]
Does this mean other providers API prices for Deepseek models will also increase?
forsalebypwner 6 hours ago [-]
Not necessarily, but could happen if the new Pro version is released and costs more to run
apercu 6 hours ago [-]
One thing I struggle with is the value proposition. One day on a specific task I'll feel like the model saved me a couple hours of work. Then days like yesterday, the model cost me half the day with context loss, repeating steps already completed, crashing/becoming unresponsive, making unpardonable mistakes in logic.
Please don't tell me I'm holding it wrong.
ygjb 6 hours ago [-]
It's not really possible to help without more information about the models you are using, the harness you are using, and how you use the harness.
You haven't even provided enough information for us to know if you are using the right tool. Saying "the model" in relation to an LLM provider that offers several is like asking for help with using a Dewalt or Milwaukee to help assemble an cabinet. Are you cutting, drilling, hammering, screwing?
apercu 3 hours ago [-]
I'm speaking pretty generally here - the tasks vary, as do the models and the way I access them. The point I was making is that they are inconsistent hour to hour, much less day to day and require a very disciplined mindset to babysit, which is sometimes harder than the actual task in play.
They do "unstuck" me sometimes. There is value in that. But the value is inconsistent - which was my original point.
ygjb 45 minutes ago [-]
Ah yeah, but that's not an AI problem in general. The models are trained on human language and human centered expression. I realized after re-reading, that my comment was more harsh than I intended, I really was trying to be helpful. The first step of asking for help should be asking if you have framed the question or task correctly, that way you build your own understanding of what you are asking the model to help you with. For better or worse, under the hood, the computer is a machine following instructions that largely appear non-deterministic, so if you don't frame the inference request correctly, it just boils down to a GIGO problem.
It's probably not what you want to hear, but are you sure you were holding the tool correctly? At least with this tool, you can actually ask it :D
WhereIsTheTruth 6 hours ago [-]
- The Launch: DS 4 officially releases
- The Big Promise: Announces a major price cut for Summer 2026, powered by cost savings from a shift to Huawei chips
- The Hype Train: Launches a flash promotion cutting prices in half, follows up by announcing the 50% discount is now permanent
- The Reality Check: Summer arrives, and Huawei's chips flops
- The Retreat: Forced to quietly reorder Nvidia chips
- The Damage Control: Introduces "off peak hours" pricing to shift demand
- The Aftermath: Announces a significant price hike
Any suggestions for a better configuration? I mostly need Opus 4.6 + Claude Code alternative. I don't need Fable level capabilities, I add one new feature at a time and approve the code before shipping it. Then test and open a PR.
Usually both Fable and Opus suggest dumb ideas but are good implementers once I tweak the idea, approve the code, and add unit and production tests.
I'm OK spending max $100/month on APIs, ideally with Zero data retention. I only need a few hours a day of coding. I don't want agents running all the time; I figure I can stay on top to each feature and wrap my head around the product and new suggestions as long as I don't build too much at once.
I'm still on a Claude Max $100 plan, but it's barely usable anymore—one call and I hit 20–30% of the 5h window on Opus 4.8. Opus 5 seems tuned to make messes, and Fable burns tokens for a level of capability I don't actually need.
It's similar to the situation where you ask whether you should walk or take the car to go refill the car with gas. I feel LLMs have a linearity embedded in them that prevents them from finding non-linear, smart solutions, especially when data is scarce. They can probably get it done, but with far more complex solutions, which increases the risk of ending up with a crazy complex codebase when it could have been much simpler and more elegant.
The other situation is when I spot a problem or a feature change that's needed after experiencing the product, and I ask how to change the codebase, it suggests something, I say "why not this other thing," and then we do the other thing.
Regarding Pi my position is that it is brilliant piece of software if you don't need extensions - give the model bash tool and let it do all it wants through it or use Pi as SDK for your own advanced harness or smth similar.
But Pi with extensions has two problems. First one: rather often they don't play well together. For example, if you want some adjustments A and B for the same tool and there are two extensions which do A and B, they will likely not work as expected when installed simultaneously. You could say that it can be solved by adjusting extensions or just generating your own - yes and it is the second problem. Like any piece of code you own and use, you have to maintain it. Bug here, incompatibility there and voila - you spend your precious time to work on harness instead of doing your job. Plus remember that vibecoders are not very responsible people, so Pi extensions registry is flooded by "use Pi to customize Pi" buggy one shot extensions.
With carefully developed set of extensions Pi would be better, like properly configured Arch Linux could be better than Linux Mint in the hands of power user. But considering how fast things are changing in this sphere, seems it is more optimal to take more bloated harness - with unneeded tools, too big prompts, etc - which will be effective on 90%, but do the actual job with that harness right now.
Not sure about the OAi Pro plan, doesn’t look like the 80% Luna price slash made its way into the quota system.
You could also try tuning down the effort level on Opus. It makes a huge difference in token consumption and you might be able to get away with lower than you’ve set
MiniMax's "token plan" ($20/mo for 1.7b tokens) is cost competitive. MiniMax M3 is equally good, if not better than DeepSeek v4, at coding: https://platform.minimax.io/subscribe/token-plan?tab=individ...
If you prefer pay-as-you-go, then Xiaomi MiMo is the only other provider with comparable models (MiMo v2.5 & Pro) that matches DeepSeek's current API rates for input/output/cache: https://mimo.mi.com/docs/price/pay-as-you-go
Meanwhile, Meta is running a 10x discount on Muse Spark 1.2 (Grok 4.5 / Sonnet 5 level model), if you opt-in to data sharing: https://dev.meta.ai/docs/getting-started/pricing-rate-limits
> I'm OK spending max $100/month on APIs, ideally with Zero data retention.
In that case, probably you'll get more out of OpenAI's coding plan, as (from what I hear routinely) the GPT 5.6 series is thrifty with token use but as smart as the Claude 5 series: https://x.com/ArtificialAnlys/status/2085083490056589784 / https://archive.vn/3VDlN
While it won’t cover everything DeepSeek does, it handles sophisticated tasks quite well. I found it after spotting it on a chart of different models and it was listed as being near Deepseek v4 Flash’s price/performance levels.
I have noticed by the way that DeepSeek’s API has been pretty slow the past couple of days. This feels like a demand-driven move more than anything. Good thing I invested in an eGPU!
Curious: Which one?
> MiMo-V2.5-Pro to be a pretty cost-effective alternative ... I found it after spotting it on a chart of different models and it was listed as being near DeepSeek v4 Flash's price/performance levels.
MiMo v2.5 Pro is at DeepSeek v4 Pro price level (but consumes lesser tokens per task, so cheaper overall). DeepSeek v4 Flash costs ~3x lesser than the Pro variant!
https://tedium.co/2026/06/10/gigabyte-aorus-5060-ti-ai-box-e...
> We plan to raise the overall pricing for DeepSeek API services in the near future, with a significant increase expected. Please plan your usage accordingly. The specific pricing plan will be subject to official notice.
I guess this means that this powerful model for very cheap concept does not work?
Their v4 flash is brilliant, and I expect they have had a huge, and expensive to handle at short notice, surge in customers as a result.
Nearly irrelevant for US/Eastern unless you’ve got major insomnia.
Remember how cheap Uber rides were? That being said, current models are absolutely incredible. And I’ll think they’re going to get cheaper when the next generation is released.
The good times never last.
We plan to raise the overall pricing for DeepSeek API services in the near future, with a significant increase expected. Please plan your usage accordingly. The specific pricing plan will be subject to official notice.
but compared to the $500k+ racks running the current frontier, it does give perspective.
It would be nice to have this notice on the public documentation as well.
If not, the cost outweighs whatever value I might have gotten from it.
I put 10 bucks on DeepSeek almost 3 months ago.
I still have 2 bucks there. I think I used so far something in the vicinity of 300M tokens total.
Even if they double their price it is still cheaper than US models on their flat rate plans, and I don't get locked out for expiring the quota.
> For us, a reasonable profit means roughly this: we buy a batch of servers, and we recover the cost in about ten months. Given the risks and the upfront investment, even if we depreciate a server financially over three or five years, commercially we think a ten-month payback is enough. That is the logic behind our current API pricing. For V3.2 Flash and other models, the standard is the same: recover the cost of the equipment in ten months.
[0]: https://thechatr.ai/blog/deepseek-liang-wenfeng-investor-mee...
We plan to raise the overall pricing for DeepSeek API services in the near future, with a significant increase expected. Please plan your usage accordingly. The specific pricing plan will be subject to official notice.
Now, the question remains, what does that "significant" mean? Would it still be cheaper than the competition, or did they realize they are too cheap for what they offer?
Given that many inference providers offer DS4 flash for more or less the same input and output token price as DeepSeek, they have good profits even with todays low prices.
Please don't tell me I'm holding it wrong.
You haven't even provided enough information for us to know if you are using the right tool. Saying "the model" in relation to an LLM provider that offers several is like asking for help with using a Dewalt or Milwaukee to help assemble an cabinet. Are you cutting, drilling, hammering, screwing?
They do "unstuck" me sometimes. There is value in that. But the value is inconsistent - which was my original point.
It's probably not what you want to hear, but are you sure you were holding the tool correctly? At least with this tool, you can actually ask it :D
- The Big Promise: Announces a major price cut for Summer 2026, powered by cost savings from a shift to Huawei chips
- The Hype Train: Launches a flash promotion cutting prices in half, follows up by announcing the 50% discount is now permanent
- The Reality Check: Summer arrives, and Huawei's chips flops
- The Retreat: Forced to quietly reorder Nvidia chips
- The Damage Control: Introduces "off peak hours" pricing to shift demand
- The Aftermath: Announces a significant price hike
Ready to IPO :^)