2 ms·

> Kimi K2.5 is absolutely competitive for coding. Kimi K2.5 is good, but it's still behind the main models like Claude's offerings and GPT-5.2. Yes, I know wha

> Kimi K2.5 is absolutely competitive for coding. Kimi K2.5 is good, but it's still behind the main models like Claude's offerings and GPT-5.2. Yes, I know what the benchmarks say, but the benchmarks for open weight models have been overpromising for a long time and Kimi K2.5 is no exception. Kimi K2.5 is also not something you can easily run locally without investing $5-10K or more. There are hosted options you can pay for, but like the parent commenter observed: By the time you're pinching pennies on LLM costs, what are you even achieving? I could see how it could make sense for students or people who aren't doing this professionally, but anyone doing this professionally really should skip straight to the best models available. Unless you're billing hourly and looking for excuses to generate more work I guess?

deaux 9mo ago
I disagree, based on having used it extensively over the last week. I find it to be at least as strong as Sonnet 4.5 and 5.2-Codex on the majority of tasks, often better. Note that even among the big 3, each of them has a domain where they're better than the other two. It's not better than Codex (x-)high at debugging non-UI code - but neither is Opus or Gemini. It's not better than Gemini at UI design - but neither is Opus or Codex. It's not better than Opus at tool usage and delegation - but neither is Gemini or Codex.
- ianlevesque 9mo ago
  Yeah Kimi-K2.5 is the first open weights model that actually feels competitive with the closed models, and I've tried a lot of them now.
  - deaux 9mo ago
    Same, I'm still not sure where it shines though. In each of the three big domains I named, the respective top performing closed model still seems to have the edge. That keeps me from reaching for it more often. Fantastic all-rounder for sure.
- VladVladikoff 9mo ago
  What hardware are you running it on?
  - deaux 9mo ago
    I'm not running it locally, just using cloud inference. The people I know who do use RTX 6000s, picking the quant based on how many of them they've got. Chained M3 ultra setups are fine to play around with but too slow for actual use as a dev.
triage8004 9mo ago
Disagree it's behind gpt top models. It's just slightly behind opus

deaux 9mo ago

I disagree, based on having used it extensively over the last week. I find it to be at least as strong as Sonnet 4.5 and 5.2-Codex on the majority of tasks, often better. Note that even among the big 3, each of them has a domain where they're better than the other two. It's not better than Codex (x-)high at debugging non-UI code - but neither is Opus or Gemini. It's not better than Gemini at UI design - but neither is Opus or Codex. It's not better than Opus at tool usage and delegation - but neither is Gemini or Codex.

ianlevesque 9mo ago
Yeah Kimi-K2.5 is the first open weights model that actually feels competitive with the closed models, and I've tried a lot of them now.
- deaux 9mo ago
  Same, I'm still not sure where it shines though. In each of the three big domains I named, the respective top performing closed model still seems to have the edge. That keeps me from reaching for it more often. Fantastic all-rounder for sure.
VladVladikoff 9mo ago
What hardware are you running it on?
- deaux 9mo ago
  I'm not running it locally, just using cloud inference. The people I know who do use RTX 6000s, picking the quant based on how many of them they've got. Chained M3 ultra setups are fine to play around with but too slow for actual use as a dev.

triage8004 9mo ago

Disagree it's behind gpt top models. It's just slightly behind opus