Kimi K3-256k • Hacker News Board

Re: Kimi K3-256k

#151

Post by gosub100 » Thu, Jul 30, 2026, 1:08 PM UTC

adamtaylor_13 wrote: post #148

johnbarron wrote: post #145

Earlier quoted context omitted.

64k ought to be enough for anybody

Why stop there? No true developer needs more than 32k! -- Jesting aside, I cannot think of a single session in recent memory that used less than ~70k tokens, so I presume you are joking.

It was from a famous quote supposedly from bill gates in the early days of computing: "640K (RAM) ought to be enough for anybody".

Apparently it was taken out of context or misattributed or something. Similar to "Al Gore said he invented the internet"

Re: Kimi K3-256k

#152

Post by jingpostmedia » Thu, Jul 30, 2026, 1:14 PM UTC

holoduke wrote: post #31

A bit of topic. But how likely is it that the US will restrict Chinese open weight models and also force Euro countries to do the same? I think it will be effective within 6 months. The US is having a hard time staying competitive.

[flagged]

Re: Kimi K3-256k

#153

Post by Roark66 » Thu, Jul 30, 2026, 1:41 PM UTC

This is very cool that they open weight it. Are there any EU providers serving it yet?

Re: Kimi K3-256k

#154

Post by Roark66 » Thu, Jul 30, 2026, 1:45 PM UTC

verdverm wrote: post #101

okrad wrote: post #61

Earlier quoted context omitted.

I believe Codex harness is pretty sticky though I haven’t tried many others. Does anyone else provide a harness of that quality?

opencode is highly regarded, I use it exclusively

Open code is cool once I added the ctrl-o function to it (show thinking and command outputs at will instead of on by default), but sadly I don't think it got merged by the team.

I forgot why.

But I can't use an AI cli without that feature.

Re: Kimi K3-256k

#155

Post by nvarsj » Thu, Jul 30, 2026, 1:56 PM UTC

yeswecatan wrote: post #96

nvarsj wrote: post #73

Earlier quoted context omitted.

Pi harness with subagents is pretty great. I’m using it for everything now.

How are you handling subagents?

tintinweb/pi-subagents. Then I use superpowers and it spins up subagents automatically. Works really well and you can customise the models and efforts you want to use with local AGENTS.md.

I’m primarily using gpt-5.6 and then opus for reviews.

Re: Kimi K3-256k

#156

Post by denis-c » Thu, Jul 30, 2026, 1:58 PM UTC

Roark66 wrote: post #153

This is very cool that they open weight it. Are there any EU providers serving it yet?

We are trying https://tensorx.ai/ as they have good model selection and ZDR, but seem new so we will see. Performance so far is good.

Re: Kimi K3-256k

#157

Post by pornel » Thu, Jul 30, 2026, 2:35 PM UTC

tripplyons wrote: post #144

pornel wrote: post #104

Earlier quoted context omitted.

RAM needed for keeping KV cache around may be the more expensive factor.

You only need to cache for tokens that are actually being used. Using 100k tokens of a 256k token window takes the amount of memory as using 100k tokens of a 1m token window.

which is why they charge extra for an option to hold >256K tokens.

The cost may be smoothly variable, but likely there's a bimodal distribution of users who barely use any context and users who push it to the max. Average price across both extremes fits nobody, but averages per kind of workload can be close enough.

Having it as a separate model makes it easier to load-balance the traffic.

Re: Kimi K3-256k

#158

Post by verdverm » Thu, Jul 30, 2026, 3:39 PM UTC

Roark66 wrote: post #154

verdverm wrote: post #101

Earlier quoted context omitted.

opencode is highly regarded, I use it exclusively

Open code is cool once I added the ctrl-o function to it (show thinking and command outputs at will instead of on by default), but sadly I don't think it got merged by the team. I forgot why. But I can't use an AI cli without that feature.

you can turn those on through the existing menu system

Re: Kimi K3-256k

#159

Post by verdverm » Thu, Jul 30, 2026, 3:40 PM UTC

versteegen wrote: post #131

verdverm wrote: post #101

Earlier quoted context omitted.

opencode is highly regarded, I use it exclusively

...but we're talking about compaction, and opencode's compaction is (or was) terrible. I've seen so many horrible problems that I keep it disabled (with an envvar flag, because even the config flag to turn it off was broken).

I don't use compaction, I keep my context limited, my sessions fresh, and have agents write markdown files as needed

Re: Kimi K3-256k

#160

Post by L-patpat » Thu, Jul 30, 2026, 4:27 PM UTC

When I first built features with GLM, there were lots of bugs, and it took me ages to fix them manually. Now the features implemented with GLM 2.0 have almost no critical bugs after testing. I can’t even imagine how capable K3 will be. It may well be on a level that ordinary people cannot access.