I've been using it with opencode. You can either use your kimi code subscription (flat fee), moonshot.ai api key (per token) or openrouter to access it. OpenCode works beautifully with the model.
Edit: as a side note, I only installed opencode to try this model and I gotta say it is pretty good. Did not think it'd be as good as claude code but its just fine. Been using it with codex too.
I tried to use opencode for kimi k2.5 too but recently they changed their pricing from 200 tool requests/5 hour to token based pricing.
I can only speak from the tool request based but for some reason anecdotally opencode took like 10 requests in like 3-4 minutes where Kimi cli took 2-3
So I personally like/stick with the kimi cli for kimi coding. I haven't tested it out again with OpenAI with teh new token based pricing but I do think that opencode might add more token issue.
Kimi Cli's pretty good too imo. You should check it out!
I was using it for multi-hour tasks scripted via an self-written orchestrator on a small VM and ended up switching away from it because it would run slower and slower over time.
Running it via https://platform.moonshot.ai -- using OpenCode. They have super cheap monthly plans at kimi.com too, but I'm not using it because I already have codex and claude monthly plans.
> Doesn't list Kimi 2.5 and seems to be chat-only, not API, correct?
Yes, it is chat only, but that list is out of date - Kimi 2.5 (with or without reasoning) is available, as are ChatGPT 5.2, Gemini 3 Pro (Preview), etc
> The 1.8-bit (UD-TQ1_0) quant will run on a single 24GB GPU if you offload all MoE layers to system RAM (or a fast SSD). With ~256GB RAM, expect ~10 tokens/s. The full Kimi K2.5 model is 630GB and typically requires at least 4× H200 GPUs.
If the model fits, you will get >40 tokens/s when using a B200.
To run the model in near full precision, you can use the 4-bit or 5-bit quants. You can use any higher just to be safe.
For strong performance, aim for >240GB of unified memory (or combined RAM+VRAM) to reach 10+ tokens/s. If you’re below that, it'll work but speed will drop (llama.cpp can still run via mmap/disk offload) and may fall from ~10 tokens/s to <2 token/s.
We recommend UD-Q2_K_XL (375GB) as a good size/quality balance. Best rule of thumb: RAM+VRAM ≈ the quant size; otherwise it’ll still work, just slower due to offloading.
I'm running the Q4_K_M quant on a xeon with 7x A4000s and I'm getting about 8 tok/s with small context (16k). I need to do more tuning, I think I can get more out of it, but it's never gonna be fast on this suboptimal machine.
you can add 1 more GPU so you can take advantage of tensor parallel. I get the same speed with 5 3090's with most of the model on 2400mhz ddr4 ram, 8.5tk almost constant. I don't really do agents but chat, and it holds up to 64k.
That is a very good point and I would love to do it, but I built this machine in a desktop case and the motherboard has seven slots. I did a custom water cooling manifold just to make it work with all the cards.
I'm trying to figure out how to add another card on a riser hanging off a slimsas port, or maybe I could turn the bottom slot into two vertical slots.. the case (fractal meshify 2 xl) has room for a vertical mounted card that wouldn't interfere with the others, but I'd need to make a custom riser with two slots on it to make it work. I dunno, it's possible!
I also have an RTX Pro 6000 Blackwell and an RTX 5000 Ada.. I'd be better off pulling all the A7000s and throwing both of those cards in this machine, but then I wouldn't have anything for my desktop. Decisions, decisions!
I tried kimi k2.5 and first I didn't really like it. I was critical of it but then I started liking it. Also, the model has kind of replaced how I use chatgpt too & I really love kimi 2.5 the most right now (although gemini models come close too)
To be honest, I do feel like kimi k2.5 is the best open source model. It's not the best model itself right now tho but its really price performant and for many use cases might be nice depending on it.
It might not be the completely SOTA that people say but it comes pretty close and its open source and I trust the open source part because I feel like other providers can also run it and just about a lot of other things too (also considering that iirc chatgpt recently slashed some old models)
I really appreciate kimi for still open sourcing their complete SOTA and then releasing some research papers on top of them unlike Qwen which has closed source its complete SOTA.
That tends to work quite poorly because Claude Code does not use standard completions APIs. I tried it with Kimi, using litellm[proxy], and it failed in too many places.
For the signal to noise reason, I start with Claude Code reviewing a PR. Then I selectively choose what I want to bubble up to the actual review. Often times, there's additional context not available to the model or it's just nit picky.
Wait, so you have the LLM review, then you review (selectively choose) the proposed review, then you (or the LLM?) review the reviewed review? But often times (so, the majority?) the initial LLM review is useless, so you're reviewing reviews that won't pass review...
Sounds incredibly pointless. But at least you're spending those tokens your boss was forced to buy so the board can tell the investors that they've jumped on the bandwagon, hooray!
Supabase seems to be killing it. I read somewhere they are used by ~70% of YCombinator startups. I wonder how many of those eventually move to self-hosted.
I had a lot of fun reading the articles about Gas Town although I started to lose track of the odd naming. Only odd because they make sense to Steve and others who have seen the Mad Max, Water World movies.
I promptly gave Claude the text to the articles and had him rewrite using idiomatic distributed systems naming.
Thank you! Is this the future? Everyone gets to have their own cutesy translation of everything? If I want "kubectl apply" to have a Tron theme, while my coworker wants a Disney theme. Is the runbook going to be in Klingon if I'm fluent in that?
Before I clicked on this I was optimistic and thought this was going to be about how we've turned a corner and the web stack pendulum is now swinging back to the easier days before frontend frameworks.
A web app platform written in Rust with the primary focus on zero-dependency apps and using Pingora as a forward and reverse proxy. Targeting Hetzner for hosting and Cloudflare for DNS. I love Rust but don’t like the long compile times which led me down this rabbit hole (zero dependencies make for fast compiles).
For a while, the O'Reilly subscription was included in the $99/yr ACM membership. Then they stopped offering O'Reilly for a bit. Then they brought it back as part of the $75 skills add-on.
I feel like this is a little known secret (discount via ACM) that more folks should know about. Hopefully this post helps spread the word.
reply