frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Kimi K3-256k

https://www.kimi.com/code/docs/en/kimi-code/models
114•monneyboi•54m ago

Comments

madihaa•41m ago
That's actually nice! I usually try to stay below 200k context anyway.
giancarlostoro•38m ago
For me the sweet spot is somewhere under 500k depending on how extensive I want to get. You can build up a sizable effort project in half a million tokens with Claude, with Claude having all the context from ground 0 to wherever you're off at.
cyanydeez•33m ago
I'm always curious what you guys are working on; every git repo I've run a local model on and stick below <100k to increase speed seems effective enough to scope patches and changes.
jdoe1337halo•11m ago
They are just talking to the model in CC, while staying in a single thread. Doubt they have any actual coding knowledge to compartmentalize different problems in the codebase.
KronisLV•9m ago
My current Claude Code session has been going on for like 35 hours and has used up around 400 million tokens, thankfully almost all of those being cached (95-98%) - pretty typical for long form agentic work.

First you spend like 2-3 hours working on a plan, once you have that you just tell the model to go and implement it, do adversarial sub-agent review loops before each commit and also make sure that all tooling and tests pass (including coverage requirements). You do need to poke it in a slightly different direction every few hours, though. Not even any novel work, just some refactoring and SSE notification hardening, bug fixes, alongside environment tuning and getting rid of some bottlenecks (also migrated from Oracle to PostgreSQL but that's mostly done).

That said, Kimi somehow manages to use less context in the main thread than Anthropic's models (even when you use sub-agents and also dynamic workflows in Claude Code), might have something to do with either how the model is tuned or their Kimi Code harness - because even in most of the longer form sessions it doesn't seem to fill up quite as quickly (note: because the kimi vis tool doesn't have a full summary view across all agents, these are the main long running agent stats across some sessions, not sub-agents):

  total tokens    cache hit rate    wall time    peak context
  283M            98%               3963m        466k
  258M            97%               2724m        467k
  98M             94%               1353m        393k
  67M             97%               614m         434k
  75M             98%               1447m        498k
  53M             99%               191m         375k
  6M              96%               139m         124k
  7M              98%               86m          118k
  11M             99%               61m          147k
I could see 256k context being sufficient for all sorts of work, even if intermediate progress/plan tracking files and docs might have to be used along the way, in addition to whatever plan support the harness has (for example, if you document something that will be relevant for load testing you might need that in 10 turns but not during the ones before then).
superloika•40m ago
The bells tolled today, but nobody came to church.
hawtads•34m ago
This is just an API level change right? The model itself should be the same I think.
wxw•27m ago
> k3-256k is now available. Within 256k context, it delivers the same results. k3 (1M) consumes about twice as much quota as k3-256k.
dgritsko•26m ago
This isn't quantized, right? Just a smaller context?
DSingularity•16m ago
Its 256k context window. Quantization is orthogonal. We cant really tell directly so it could be quantized.
ibuildproducts•23m ago
omg! new model!!
lukan•13m ago
Since Claude is the first time for me really, really out (TIL against my wished about https://status.claude.com/), I am now interested enough to see what else works. But ... when I click pricing, I see "Join a waitlist". Wtf? Are they really that good, so were totally surprised and overwhelmed by the requests, is this a marketing stunt, or do they just don't have the hardware being in china?
MYEUHD•8m ago
Since the model is open-weights, you can get from other providers, for example see https://openrouter.ai/moonshotai/kimi-k3#providers
KronisLV•2m ago
> Are they really that good,

They did exceed the expecations of pretty much everyone! I've also blogged about using the model, it's a bit on the slow side but pretty good!

> so were totally surprised and overwhelmed by the requests ... or do they just don't have the hardware being in china?

Yes, this is mostly the case: https://x.com/Kimi_Moonshot/status/2078855608565207130

As a user, I much prefer that to service disruptions or severely degraded or secretly quantized performance. However if I didn't have an account, I'd be pretty pissed off about not being able to give them money and become a user.

illithid0•13m ago
This was posted 38 minutes ago, and as of 20 minutes ago, several Anthropic services are now designated as having a "major outage".

Doubt these are related, but it made me laugh a little.

RimWorld's Code Defines Strict Gender Roles (2016)

https://www.rockpapershotgun.com/rimworld-code-analysis
1•Tomte•41s ago•0 comments

HAR Extractor

https://github.com/azu/har-extractor
1•soupspaces•2m ago•0 comments

Show HN: Resume Tailor – Fit your resume to any job

https://tailor.resume.academy/
1•vednig•3m ago•0 comments

A Trampoline

https://dogdogfish.com/blog/2026/07/29/a-trampoline/
2•matthewsharpe3•5m ago•0 comments

Basebrowserproject

https://mastodon.social/@sarahjamielewis/116965925134524215
1•birdculture•5m ago•0 comments

XGBoost 3.3.0 Release

https://xgboost.ai/2026/07/21/xgboost-3-3-release.html
2•nickbecker•6m ago•0 comments

Anyone else think Claude Code and Codex are too slow? We hate it

https://bullet.davidhf.com
3•alsima•7m ago•0 comments

YC startup offers interviews to people who get their name permanently tattooed

2•codexon•10m ago•0 comments

Neuroprosthesis restores hand movement&sensation in human w complete tetraplegia

https://www.nature.com/articles/s41591-026-04498-0
1•bookofjoe•11m ago•0 comments

Benchmarking LLMs on SAST Triage

https://www.fencer.dev/blog/llm-triage-sast-false-positives
5•timols•11m ago•0 comments

FrameCut – Split contact sheet negative film scans in Chrome

https://github.com/stabruriss/framecut/blob/main/README.md
1•chriswunan•11m ago•0 comments

Runlayer sues Rippling, alleging source-code access preceded rival AI gateway

https://runtimewire.com/article/runlayer-sues-rippling-alleging-source-code-access-preceded-rival...
2•ryanmerket•12m ago•0 comments

Collection of Vintage User-Interfaces

https://typewritten.org/Media/
1•karakoram•12m ago•0 comments

A digitally controlled silicon quantum processing unit

https://www.nature.com/articles/s41586-026-10754-7
1•sbulaev•12m ago•0 comments

China's Moonshot AI reportedly used Nvidia Blackwell chips for training Kimi K3

https://www.tomshardware.com/tech-industry/artificial-intelligence/chinas-moonshot-ai-reportedly-...
2•rndsignals•14m ago•0 comments

Thorpe – semi-WYSIWYG Markdown editor

https://thorpeapp.com/
1•jasoncartwright•15m ago•0 comments

Botswana Did Everything Right

https://www.cremieux.xyz/p/botswana-did-everything-right
1•admp•15m ago•0 comments

Microsoft FY26 Q4 Earnings

https://www.microsoft.com/en-us/Investor/earnings/FY-2026-Q4/press-release-webcast
3•corvad•16m ago•1 comments

Meta Reports Second Quarter 2026 Results

https://investor.atmeta.com/investor-news/press-release-details/2026/Meta-Reports-Second-Quarter-...
1•mfiguiere•16m ago•0 comments

Kimi k3 now runs on one consumer GPU

https://twitter.com/Akashi203/status/2082555972380401852
5•OsamaJaber•17m ago•0 comments

Why too much time online makes us unhappy

https://www.cnn.com/2026/07/29/health/too-much-time-online-wellness
2•01-_-•18m ago•1 comments

Schools Are Confiscating Phones All Day

https://www.gadgetreview.com/schools-are-confiscating-phones-all-day-and-most-americans-are-fine-...
1•01-_-•20m ago•0 comments

Cory Doctorow on Why AI Won't Replace Workers, but Will Crash the Economy [video]

https://www.youtube.com/watch?v=rRRmUuxJolY
1•SockThief•20m ago•0 comments

The GitHub Platform Era Is Over

https://julien.danjou.info/blog/the-github-platform-era-is-over/
1•yruzin•21m ago•0 comments

The Prevention Paradox: Why Successful Resilience Work Becomes Its Own Enemy

https://newsletter.resiliumlabs.com/p/the-prevention-paradox
1•gpi•23m ago•0 comments

Cloudflare removes top open coding models from free tier

https://developers.cloudflare.com/changelog/post/2026-07-28-models-require-workers-paid/
1•ankushKun•24m ago•1 comments

Kimi K3 for local use (1.56TB –> 594GB) compressed by Unsloth

https://huggingface.co/unsloth/Kimi-K3-GGUF
4•pavelai•25m ago•1 comments

Can You Survive the Kobayashi Maru?

https://moai.studio/blog/posts/rewrite-the-simulator.html
1•ionwake•26m ago•0 comments

Why are databases so hard?

http://gtowey.blogspot.com/2026/07/why-are-databases-so-hard.html
1•gtowey•27m ago•0 comments

Personal Instagram Clone: WordPress and ActivityPub

https://fotos.kaffeeringe.de/
1•kaffeeringe•27m ago•1 comments