#Task - cost

1 messages · Page 1 of 1 (latest)

void delta
#

I have observed an inconsistency in the credit consumption. While I understand that the expanding context window as a project progresses might justify a higher cost per task, the current model lacks predictability and seems disproportionate.
In contrast, tools like Copilot Pro maintain a fixed cost of one credit per interaction, even if the generation process extends for several minutes, without this increasing the price. This predictability, combined with a significantly superior performance, highlights a gap in the current value proposition. Although I understand the product is in a "beta" phase, I suggest that the cost model be reviewed to better align it with the value and performance delivered.

PD: To provide context for my feedback, I have benchmarked the performance of initial tasks on your tool against Copilot Pro and Sonnet 3.5, and the results suggest that your value proposition should be re-evaluated. My sole intention is to offer a constructive perspective aimed at improving the product. I recognize the significant development effort behind such a complex tool, and my interest in thoroughly testing it stems from its clear potential.
Furthermore, I have identified opportunities for improvement in error handling. In the first version I tested a few months ago, I experienced errors when moving or copying files. This time, an error occurred when I tried to upload assets that exceeded GitHub's maximum file size limit. After three failed iterations, I had to manually intervene by instructing it to add the folder with the large files (assets/models) to the .gitignore and start over. In comparison, tools like Copilot Pro also encounter errors, but they demonstrate a greater ability to diagnose and resolve them autonomously.

void delta
#

Lately, it sometimes gets stuck for minutes on end doing nothing. I assume that after 10 minutes, it's not actually processing anything, so you have to cancel and retry several times. It also gets stuck on testing errors; after several iterations, it decides to resolve them using mock data. I think a better approach would be to have the necessary mock data for testing prepared in advance, so the code doesn't have to be modified later to work with real data. It does some unpleasant things.
The solution isn't to create a simpler test by removing the ones that don't work. You should inform the user that the task has been completed but that these 4 tests failed for specific reasons. Then, they can decide whether to fix them or move on to other tasks.

bright salmon
#

I don't understand the pricing is fair if you ask 'Give the answer to the question A' and the answer is 'A=23' for this response Copilot will use one full credit and Kiro can use 0.1 so you can ask 10 times the same thing to Kiro for the same token usage of Copilot

#

About you getting stuck yeah it happen, what i have done was to tell in my prompt to give the precise hour when it does a command and i keep a watch

#

for let's say a 'npm run dev' that get stuck on the '[nodemon] error...' i create a script to capture it and put an end to the terminal process, this way i can say 'with context7 and sequential-thinking mcp i want you to do npm run dev:capture until the server start, if there's an error captured then follow the steps given with the error' and the prompts are basicaly 1- You create a plan of actions with mcp sequential-thinking that include researching the specific documentation with context7 MCP / 2- You do that follow this rule and go on..

#

basically you need to work around the limitations, as you said Kiro is still in Beta

void delta
#

In Copilot, you get 4 free models and another 3 at a 0.33x rate. It doesn't show me decimals for the usage counts, but I can tell you that I was just charged 5 credits in the span of 2 minutes. On top of that, they also charge credits for generating a final summary of the implementations. I was even charged two credits once because it apparently got stuck while reading the task list and the implementation file.
We'll have to wait for improvements and give it one last chance. I like to support new projects so that the established players aren't the only ones left, because a lack of competition is never a good thing.

#

not return credits...

bright salmon
#

The credits usage you are seeing in your tchat isn't 100% your credit usage, it's only an indication, a prediction of what you have consumed

#

They can't tell you for sure unless they wait a day or somes because it they fetched all request of everybody at every moment, they won't be able to keep up

#

so until the big data transfer that happen daily (depending on the service) they tell you more or less what you have consumed!

#

So when a request is canceled and you see your credit usage being consumed, it's not true, your credit will be back at any given moment without you noticing

#

Also about Copilot there's no free model i get rate limited every single day and have to wait the next day to be able to use it again, Copilot is not at all transparent of the tokens limit so they can do whatever they want

#

with Kiro at least i can predict how much tokens i have left so i already found that better

#

also don't forget that Kiro is 3 months old where Copilot have more then a year 😉

#

Also i don't know wtf you are speaking about when you say "a lack of competition is never a good thing" there's a literal war between ADE (A for agents) every 2 days you have a new one that come out and is a tiny much better than the previous one (ex: Kilo Code -> Kiro Code)

#

You are free to believe what you want but i find you a little unfair with Kiro, it's really better than you think but you have to use MCP, you have to contextualise your project, and go on

void delta
#

Right now, I'm stuck on a Flutter test, which is ironic because all I really want to do is create a visually appealing app prototype.
Speaking of tools, I'm not entirely sure what the usage limit is for Copilot Pro, which costs $10. I get it for free as a student, but for that price, you get access to several models at no extra cost. Although I don't know the exact limit, I use GPT-4.1 for simple tasks like generating outlines and patterns, and I use GPT-4o for internationalization because it produces excellent translations. It also includes free access to GPT-5 Mini and Grok Mini, but I don't use those.
I've actually never hit the limit on the free tier. As for the premium plan, the $10 gets you 300 requests. To give you some perspective, with Kiro, you'd get 600 requests for what it costs (implying double the price). However, with Kiro, I can burn through all my requests in a single intense workday, which would be impossible for me to do with Copilot Pro.
One of Copilot's biggest strengths is its efficiency. In a single request to a model like Claude 3 Sonnet, I can have it handle the internationalization for an app across multiple languages. If it exceeds the context limit, it automatically opens a new window with a summary of the previous conversation to maintain context, and all of this consumes only a single credit.
I'm just mentioning this in case you weren't aware. That said, I completely understand the level of development and effort behind each platform. It's clear that Copilot has a mature team and over a year of development behind it, whereas Kiro, despite being backed by AWS, just isn't as mature yet.
I have to say, Kiro's workflow really impressed me when I first tried it a few months ago, even though it made a lot more mistakes back then. I'll have to see if I can test it again with its newer models, but no matter how good it is, it's not sustainable if the credits are gone in just a couple of days. I'll need to figure out a solution for that.

#

By the way, I'll give that tool you mentioned, Kilo Code, a try. Thanks a lot for the reply. For now, I'll stick with Kiro, but I'm cutting it close—I have fewer than 100 credits left for about 30 remaining tasks, so I don't think it'll be enough to finish the job.

void delta
void delta
# bright salmon The credits usage you are seeing in your tchat isn't 100% your credit usage, it'...

Hi, are you on the Kiro team?
I have a question about the accounts. I've seen some users who shared their screen with the $20 plan, and they have 1000 credits plus an additional 1000 promotional credits. My question is, if I sign up for that same plan, will I receive them as well? Right now, on my free account (which I hadn't used in months), I only received 500 promotional credits.
On another note, as you rightly mentioned, for now, it's necessary to connect it via MCP to the console until they fix it. Related to this, I've discovered a potential issue: when you mark a command as "trusted," sometimes it executes without prompting, and other times it doesn't. However, if you run the command without adding it to the trusted list, it always works correctly.
I'm not sure if it would be helpful for me to send this information to them, but I assume they are already aware of these issues.

hazy verge