Kimi API Platform

“As many have pointed out, Kimi K3 and K3 256K are great.However, the quota given is very bad. It's not that it refreshes every 5 hours and 7 days and that's it — rather, you can literally use your monthly quota within a day if you're not careful. I subscribed to Kimi because I need the multimodal model.A bonus when K3 came out: K3 has a 1 M token context window, which is great and, in fact, useful. The Kimi model itself is multimodal, which I needed because the GLM coding plan does not include v”

“I tested Kimi and it was, by far, the best all-rounder ai agent I had used up till now. So, I upgraded to the Allegro plan.was...At the moment it's all frustration with kimi...to the point I manually moved back to chatgpt.It can't remember what was said in the previous line. Doesn't save the memories I tell it to. CONSTANTLY hallucinates. constantly outputs without me asking it to. constantly contradicts itself from one reply to the next.I initially was in love with the claw service. But then, i”

“- Consume way more tokens than Claude (so it's not cheaper) - It's fucking slow. - It's coming empty, you have to configure it. - Aggressive marketing campaign with YouTubers, with content controls on the "Claude Killer" topics. Omg guys, try it and you'll see. It's not that bad, it's just not worth it. This is just a cheap copy. Don't take the bait.”

“Every time I read these I just assume you don't know how to prompt and you don't know how to set up your dev environment. You guys do know that K3 is not meant to be lightweight, right? You can use Kimi 2.7 for days on end but K3 burns tokens similarly to Opus and Fable. So you have the smart model plan and you have lesser agents do work. You'll get better results and never run out of tokens like you are now.”

“I switched my workflow over to k3 via openrouter and found it to be less than half the cost of opus 4.8 but getting similar, if not better results. I came to the subreddit expecting to find others talking about this insane price difference, but instead everyone here is saying how much more expensive it is than Claude. The difference being, every post here is about the yearly plan with quotas and limits. Per token the API on paper should be only about 40% cheaper than Opus, but it spends less tim”

“the price comparsion in the sub is usually coding plans to coding plans, so that's just what the people use. api prices break my brain on anything that's mimov2.5 and deepseek stuff. Used gpt 5.4 on api before...got like a 700 dollar bill for like half a day of work....so um..never again”

“I'm using Kimi since March 2026. And it was awesome, I've shifted from Claude, still using Codex. Why I'm cancelling? THE LIMITS DRAINS INSANELY FAST! Even Kimi K3 256 High drain your 5h (Allegro) limit in few hours. (I'm updating Kimi regularly) Cache is NOT working. Kimi state that if you're using the thread within an hour and not have pause for more then hour in your messages you'll match the cache. During conversation I can drop a simple INFO question after 2 minutes of thinking I have -12% ”

“The cache issue would probably frustrate me more than the raw limits. A model can be amazing, but if every follow-up burns a meaningful chunk of your quota because the context isn't being cached properly, the effective price per completed task gets ugly very quickly.”

“bought max plan x20 from a reseller for basically nothing compared to official pricing. It works. No issues yet. Now I can’t stop thinking about How are they not losing money?”

“I'm using Kimi since March 2026. And it was awesome, I've shifted from Claude, still using Codex. Why I'm cancelling? THE LIMITS DRAINS INSANELY FAST! Even Kimi K3 256 High drain your 5h (Allegro) limit in few hours. (I'm updating Kimi regularly) Cache is NOT working. Kimi state that if you're using the thread within an hour and not have pause for more then hour in your messages you'll match the cache. During conversation I can drop a simple INFO question after 2 minutes of thinking I have -12% ”

“The cache issue would probably frustrate me more than the raw limits. A model can be amazing, but if every follow-up burns a meaningful chunk of your quota because the context isn't being cached properly, the effective price per completed task gets ugly very quickly.”

“Same here (well, not same – I'm using Kimi for writing and translation and Moderato). The limit hits way too fast, there is no benefit for me as compared to Claude or ChatGPT. Also Kimi is very slow. It takes minutes to do a task Claude does in seconds.”