I'm so glad and proud to be part of Kilocode. As users we can see how everything is sold in the name of transparency. Your article breaks it down really well how the whole AI credits/token/paid plan ecosystem shifts differently after a user commits to something initially and then gets trapped in predictibilty selling engines. It's difficult to stay honest and say it the way it is when there is so much competition.
I really appreciate and salute Kilocode for being usercentric first without thinking profits only. It takes some serious balls to stick with what you stated from the beginning and continue putting out the best features in the interest of the users. My heart goes out to the Kilocode team and I wish you guys/gals the best wishes for the future. I will forever be grateful and continue to be your customer no matter what...
Thanks from the bottom of my heart and I really think Kilocode will emerge against all odds. I love all the features and I expect a lot more things from Kilocode in the future.
By the way, any plans for bringing subagents or something similar in the near future? It seems every other cli is getting agent intensive.
I am sorry, but charging per token is NOT predictable and it is NOT transparent. With electricity and phone bills, people greatly prefer unlimited, unmetered or flat rate plans over paying per byte or kilowatt, because sometimes you use less, sometimes you need more, and you don’t want to have your bill balloon on the month that you use more. For most people, your usage is not consummate with your income.
Personal users don’t magically make more money when they use their LLMs more because they aren’t running it for a business. Business users who are heavily using an LLM probably have a lot of development going on, but are not making much profit in the beginning while most of the development is happening because their product is not launched.
It’s not transparent because as a user, it’s nearly impossible to predict ahead of time how many tokens will be needed to solve a particular task. If I told you to estimate the token count of adding a button to a web app, and therefore derive the cost based on the number of tokens, would you be able to tell me if that might cost $0.10, $1.00, $10.00 or $100.00? I wouldn’t, without some extremely onerous math. And what if it gets it wrong? What if it wastes a bunch of tokens doing things that are off-base? What if I want to have it passively code in an agentic loop while I play tabletop RPGs for 4 hours, and come back it used $1000 of credits?
It’s not rocket science: set up a bunch of price tiers so that the average user at each tier uses few enough resources that it’s profitable. OpenAI does it with ChatGPT Pro, which supports Codex CLI. Google does it with Gemini CLI thanks to AI Ultra. Anthropic has done it for a long time with Claude Code and Claude Max. There is a quota, but it’s so generous at the top tier that most users who aren’t abusing the system (e.g. with multiple developers working with an account meant for one developer) can both stay within the quota and afford it.
I do think Warp vastly messed up their pricing. They messed it up bad, by adding BYOK and per token pricing. What they should have done is just hiked their Lightspeed price up to by $25 or $30 per month to account for the expense of the plan, and maybe hike the starting plan by $5 per month.
I could not ever use Kilo because I’m not interested in my bill being a roller coaster. I’m fine with paying Kilo hundreds of dollars for usage I MIGHT need but don’t actually need every single month, in the hopes that when I need it, it’s there.
Ever hear of health insurance? I guess the Kilo guys like paying $100k out of pocket for major surgery in exchange for not having any insurance premium while healthy?
Predictable pricing is predictable because it’s FLAT. A price that goes up and down based on your usage is the opposite of predictable because usage is not predictable.
I'm so glad and proud to be part of Kilocode. As users we can see how everything is sold in the name of transparency. Your article breaks it down really well how the whole AI credits/token/paid plan ecosystem shifts differently after a user commits to something initially and then gets trapped in predictibilty selling engines. It's difficult to stay honest and say it the way it is when there is so much competition.
I really appreciate and salute Kilocode for being usercentric first without thinking profits only. It takes some serious balls to stick with what you stated from the beginning and continue putting out the best features in the interest of the users. My heart goes out to the Kilocode team and I wish you guys/gals the best wishes for the future. I will forever be grateful and continue to be your customer no matter what...
Thanks from the bottom of my heart and I really think Kilocode will emerge against all odds. I love all the features and I expect a lot more things from Kilocode in the future.
By the way, any plans for bringing subagents or something similar in the near future? It seems every other cli is getting agent intensive.
Indeed, this is why I choose Kilo Code, the freedom to use it and just pay for what I need when the better models are required.
And that's why I love Kilo. Transparency.
I am sorry, but charging per token is NOT predictable and it is NOT transparent. With electricity and phone bills, people greatly prefer unlimited, unmetered or flat rate plans over paying per byte or kilowatt, because sometimes you use less, sometimes you need more, and you don’t want to have your bill balloon on the month that you use more. For most people, your usage is not consummate with your income.
Personal users don’t magically make more money when they use their LLMs more because they aren’t running it for a business. Business users who are heavily using an LLM probably have a lot of development going on, but are not making much profit in the beginning while most of the development is happening because their product is not launched.
It’s not transparent because as a user, it’s nearly impossible to predict ahead of time how many tokens will be needed to solve a particular task. If I told you to estimate the token count of adding a button to a web app, and therefore derive the cost based on the number of tokens, would you be able to tell me if that might cost $0.10, $1.00, $10.00 or $100.00? I wouldn’t, without some extremely onerous math. And what if it gets it wrong? What if it wastes a bunch of tokens doing things that are off-base? What if I want to have it passively code in an agentic loop while I play tabletop RPGs for 4 hours, and come back it used $1000 of credits?
It’s not rocket science: set up a bunch of price tiers so that the average user at each tier uses few enough resources that it’s profitable. OpenAI does it with ChatGPT Pro, which supports Codex CLI. Google does it with Gemini CLI thanks to AI Ultra. Anthropic has done it for a long time with Claude Code and Claude Max. There is a quota, but it’s so generous at the top tier that most users who aren’t abusing the system (e.g. with multiple developers working with an account meant for one developer) can both stay within the quota and afford it.
I do think Warp vastly messed up their pricing. They messed it up bad, by adding BYOK and per token pricing. What they should have done is just hiked their Lightspeed price up to by $25 or $30 per month to account for the expense of the plan, and maybe hike the starting plan by $5 per month.
I could not ever use Kilo because I’m not interested in my bill being a roller coaster. I’m fine with paying Kilo hundreds of dollars for usage I MIGHT need but don’t actually need every single month, in the hopes that when I need it, it’s there.
Ever hear of health insurance? I guess the Kilo guys like paying $100k out of pocket for major surgery in exchange for not having any insurance premium while healthy?
Predictable pricing is predictable because it’s FLAT. A price that goes up and down based on your usage is the opposite of predictable because usage is not predictable.