At the right price, these model don't need to be the best, good enough will do. I think we're fast approaching good enough for most users.

▲

kouteiheika 3 hours ago | parent | next [-]

This. Here's a quick experiment I did yesterday.

I got a new $20 Claude subscription to try the new Fable model. I gave it a single prompt, and it barely finished, using up my whole session quota (it was at ~95% when it finished) and 10% of my weekly quota.

For comparison, with the Kimi Code $40 subscription I can pretty much constantly run two/three agents in parallel for the whole week, and I never run out of quota. I can blindly throw it at anything and everything without worrying about hitting the limits. (And it's not exactly a cheap model to run -- it has 1 trillion parameters!)

Is Kimi as good as Claude? Of course not. But you don't need the absolute state-of-art for most things. If I don't have exceptionally difficult tasks it makes no sense to use it. Just throw Kimi at it, and even if it needs to run 2 or 3 times longer in the background I don't care, because I'm not running out of tokens there.

▲

nl 2 hours ago | parent [-]

A word of caution on this.

I've tried this too, and was disappointed.

Kimi generally benchmarks at "a bit more intelligent than Sonnet Medium" levels[1] and I'd agree broadly with this assessment.

If you have adapted your coding to rely on the agentic style that is doable in Opus 4.7+ then you will find Kimi disappointing.

If you are using it in a more targeted way then it can work well.

[1] https://artificialanalysis.ai/agents/coding-agents?agents=cl...

▲

kouteiheika 2 hours ago | parent [-]

Yes, I would agree with this.

I think it works best when you're using the agent in a more hands-on way with a targeted prompt. If you're obsessive about code quality like I am (so you thoroughly review and, when needed, reprompt or even rewrite what the agent does) then you'll be fine, but if you like to just throw a prompt at the wall and expect it to plan and execute the whole thing perfectly then you'll be disappointed.

A middle-ground trick one can use is to have Opus (or Fable now) plan the whole thing and get something cheaper like Kimi execute on it.

	▲	rented_mule an hour ago \| parent [-]
		CodeWhale (formerly deepseek-tui) automates this over DeepSeek V4 Flash and Pro. My shallow understanding is that it prompts the model to evaluate the complexity of a given task, then decides on Flash vs. Pro at various reasoning levels for that task. This can help with both cost and speed. If other agent platforms don't already do this, I have to imagine they will at some point. I'm retired and can't justify spending too much on these things. CodeWhale over DeepSeek is helping me understand this space much better (and have some fun!), and it's quite affordable. I've spent ~30 hours using it over the last couple of weeks, and I've spent $3.89 on DeepSeek in that time. If I don't feel like writing any code for a few weeks, I pay nothing. Looking at DeepSeek's dashboard, about 60% of my requests have gone to Pro and 40% to Flash. I've used 97M Pro tokens and 19M Flash tokens (well over 90% of each have been cache hits, so the price is much lower than it would otherwise be).

▲

boc 3 hours ago | parent | prev [-]

OTOH, using the best is a competitive advantage when time = money. It's like giving your engineers a slow laptop because it's cheaper. It may be cheaper but not worth the cost.

	▲	lelanthran 2 hours ago \| parent \| next [-]
		> OTOH, using the best is a competitive advantage when time = money. It's like giving your engineers a slow laptop because it's cheaper. It may be cheaper but not worth the cost. That doesn't imply giving your devs the best laptop makes any difference. How much more productive will your devs be if you upgrade them from a 32GB RAM, 8-core laptop to a 768GB RAM 96-core threadripper? In your analogy, Kimi may not be the 4-core celeron with 4GB of RAM, it's more like the 8-core AMD with 32GB of RAM.
	▲	bushbaba 3 hours ago \| parent \| prev [-]
		Not necessarily, inference speed also has huge time aspect. For example anthropic takes nearly twice as long as OpenAI models for my tasks with both having similar success rates.