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.
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)
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?
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
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.
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'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.
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)
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.
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.
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!
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.
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.
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.
> 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.
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.
> 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.
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.
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.
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?
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.
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.
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.
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.
I'm curious what kind of prompts people are doing to get 'dumb' ideas out of Fable
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)
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?
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
I have the 20x. I really never reach the limit anymore. I am a heavy user. How is that possible?
Do you import old code base or work on small greenfield projects?
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.
> DeepSeek Flash 0731 as the implementer
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.
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”
Fireworks.ai has ZDR and hosts Kimi K3
[dead]
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)
I guess the low pricing was so DeepSeek could gather training material. The provider on OpenRouter indicates that data can be retained.
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.
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.
Link redirects to a login page.
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.
FWIW, I have found Xiaomi’s MiMo-V2.5-Pro to be a pretty cost-effective alternative. https://mimo.mi.com/models/en-US/mimo-v2.5-pro
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!
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.
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.
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.
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?
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.
You mean Beijing working time zones?
No?
> 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.
It was insane how much you could get done for a couple of dollars.
The good times never last.
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.
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.
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.
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.
The link takes me to a login page.
> 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.
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
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.
Is there currently a lot of difference between deepseek's own prices and other provider's prices of deepseek models?
Looks like other providers are around 50% higher cost https://openrouter.ai/deepseek/deepseek-v4-flash-0731#provid...
Yeah, everyone else's cache read prices are 10x higher.
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.
[0]: https://thechatr.ai/blog/deepseek-liang-wenfeng-investor-mee...
Looks like a second order effect to them not raising their round
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.
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?
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.
It’s all about the cache prices. Those dominate usage, cached tokens are easily >90% if not >95% of all used tokens.
Yes, but I'm not sure. They recently dropped cache prices 10 times. Why would they rudder back after only a few months?
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.
to fund training a far better and capable model.
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.
Does this mean other providers API prices for Deepseek models will also increase?
Not necessarily, but could happen if the new Pro version is released and costs more to run
- 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
Ready to IPO :^)
This fanfic was peak HN
[flagged]
[flagged]