[ekojs]

> In our evaluations, Kimi K3 delivers frontier-level performance. Among the models tested, its overall intelligence ranks second only to Claude Fable 5 and GPT-5.6 Sol. For the complete benchmark results, see our tech blog. The full model weights of Kimi K3 will be released in the coming days. More details on the architecture, training, and evaluation will be published together with the Kimi K3 technical report.

> K3 pushes the boundary of end-to-end knowledge work. On the GDPval-AA v2 leaderboard, Kimi K3 scores 1687. The benchmark evaluates AI models on real-world tasks across 44 occupations and 9 major industries; Kimi K3 ranks behind only Claude Fable 5 Max and GPT-5.6 Sol Max, and ahead of Claude Opus 4.8 Max at 1600.

> On AA-Briefcase, Kimi K3 scores 1527, ranking second among all models — behind only Claude Fable 5 Max and ahead of GPT-5.6 Sol Max (1495). AA-Briefcase is a private agentic knowledge-work benchmark developed by Artificial Analysis to evaluate frontier agentic capability in long-horizon knowledge work.

Really good benchmark score it seems. Maybe another DeepSeek moment right here.

[paxys]

> its overall intelligence ranks second only to Claude Fable 5 and GPT-5.6 Sol

Pretty sure ranking “second” to two others means ranking third.

[antonyt]

Charitably, you could read this as "its overall intelligence [is in a class that] ranks second only to [that of]..."

[ignoramous]

This is actually what's meant but this bikeshed has been built for yak shaving.

[stingraycharles]

Since we’re bikeshedding: that’s not what yak shaving means.

[shwaj]

Since we’re hair splitting: that’s not what bike shedding means. :-D

[esafak]

What does it look like I'm doing??

[t0mpr1c3]

yakshedding? bikeshaving? IDK

[ekojs]

Yeah, bad wording it seems. Though a charitable interpretation is that Fable 5 and GPT 5.6 Sol are joint 1st place in the measurement.

[paxys]

Doesn’t matter, the next one is still third.

[cheesecakegood]

DENSE_RANK() vs RANK() claims another victim

[UqWBcuFx6NV4r]

girl, we get it, you can count. re-stating your point does not a conversation make.

[jnwatson]

If there are two folks standing at gold, nobody gets the silver medal.

[worldthruword]

But linearizing an equal magnitude quantities by alphabet priority would be unfair. Magnitude is the important quantity here.

[jatora]

"Ranks second" is their statement. What is it's rank, in your opinion?

[novaleaf]

frontier vs "not quite" :D

[vl]

While you are technically correct, in English it’s perfectly fine to say it this way as well.

“Second only” here has meaning “next after”, not “number two”.

[light_hue_1]

Yes. "Second to" takes a set as an argument in English. Even the empty set works!

England is second to none.

[__mharrison__]

So... France took second to England and Argentina?

[vl]

France’s football team is second only to England’s and Argentina’s.

It’s a miracle that in language same words have different meanings depending on context. If this wouldn’t be the case we could have hardcoded NLP algorithmically without inventing these expensive LLMs!

[make3]

Second group essentially is how you have to think of it

[avazhi]

That’s not what second means in this context in English, and it’s incorrect to use it that way. This is because for something to be second there must have been something in first and only first, and so on; in this case there was a first and a second already, and you cannot amalgamate then because they didn’t tie (and even if they did, they’d be 1 and 2). Both logically and grammatically, it’s incorrect.

[seba_dos1]

You're both logically and grammatically wrong. You could even ask an LLM to explain the meaning of the phrase to you if you don't believe that.

[avazhi]

Either think and write for yourself or stay silent next time. It'd be infinitely better than telling another person to use an LLM to understand something you yourself don't understand and are too lazy to try to figure out.

I wish you the best.

[seba_dos1]

Hah, I had expected this knee-jerk response, but kinda hoped you'd avoid this pitfall. Alas.

See, I could tell you that in English, "second to" is a construct that usually means "next to" or "inferior to" and has nothing to do with "being in second place", and that if it did, it would make the popular construct "second only to" completely redundant. But others already did that in sibling comments before me, and you could just respond with "you're wrong" anyway, so what's the point? Pointing to an LLM is, of course, often a lazy and unhelpful cop out from the discussion, but in this particular case it's pointing you to a dataset that's explicitly about extracting meaning and finding relationships between phrases in languages - so you don't have to trust me or anyone else that this phrase is actually being used in this particular way, you can find it out yourself based on enormous training datasets illegally collected from all over the Internet.

[krackers]

Not if the others tie for first place.

[Calazon]

Still third even then.

[Demiurge]

I think what's implied here, in colloquial terms, is that it's in the second tier.

[paxys]

Except saying second tier would be bad marketing, so they decided to change the meaning of words instead.

[UqWBcuFx6NV4r]

okay, so let’s get this straight: even though there are seemingly quite a few people here that clearly understand what is being said. However, the fact that YOU specifically either genuinely do not understand this / have never come across this before, or are being intentionally difficult because of some philosophical disagreement, feel that you can unilaterally assert that they’re “redefining words”?

I don’t know if this is genuinely your first day on Earth or something, but if you’re trying to parse English like a programming language then you’re not only making things hard on yourself, but also 99% of people you’ll ever speak to.

[scotty79]

Which is still great because it means neither of the two best financed labs in the world manage to produce even two models themselves that would beat Kimi K3.

[Aurornis]

> > K3 pushes the boundary of end-to-end knowledge work. On the GDPval-AA v2 leaderboard, Kimi K3 scores 1687. The benchmark evaluates AI models on real-world tasks across 44 occupations and 9 major industries; Kimi K3 ranks behind only Claude Fable 5 Max and GPT-5.6 Sol Max, and ahead of Claude Opus 4.8 Max at 1600.

This is the same benchmark where Sonnet 5 outperforms Opus 4.8 max.

Like all model releases, the benchmarks aren't going to tell the whole story. All of the open weight models come with amazing benchmark results now. It's hard to believe anything other than that the benchmarks are leaking into (or intentionally included) into training data.

[ignoramous]

Possible, but pay-as-you-go Hy3 / DeepSeek v4 Pro / MiMo v2.5 Pro (from respective vendors) are genuinely good enough as daily drivers, given the costs (especially, low prices for input cache, which usually makes up 70%+ of total input for agentic workflows). I put in $10 in DeepSeek & Xiaomi MiMo, and I've barely used $1 each, in a week of coding work.

Coding Plans by MiniMax ($20/mo for 1.7b tokens) and Z.ai (~$30/week use for $17/mo) are also tremendous value for money.

[andai]

Sonnet 5 does beat Opus 4.8 on several benchmarks. It just costs more and takes longer.

(On several other benchmarks, it costs more, takes longer, and does worse.)

[rd]

i’ll never really understand this comment. why would labs do this if they know private benchmark evals will come out in the next week?

[adverbly]

> Maybe another DeepSeek moment right here.

Surely not... What made DeepSeek disruptive was that the cost was 10X lower.

In this case, the cost is about 2X lower the Sol I think?

At 2X, you're pretty close to the error margins due to token efficiency etc...

I'd say this is "on trend" for open models catching up to frontier labs, but its not a "change in the trend" like DeepSeek was IMO.

[avianlyric]

It was also disruptive because it was open weight, meaning anyone and their dog could theoretically compete with the frontier labs for their inference revenue.

The frontier labs need to recoup a huge amount of cash to cover their model development costs, and justify their valuations. That’s plausible when they’re only ones capable of selling inference on these models, it a lot less plausible when models themselves become cheap commodities, and you’re just competing on your ability to provide compute. Anthropic and OpenAI can’t compete with people like AWS on that front.

[verdverm]

It's different, but similar. If they release the weights, then we have a Fable / frontier model people can tinker with. Either way, it's still quite impressive and knocked a US company out of the top three (google). How long before China dominates the top-10 (if they don't already) or the #1 model?

[verdverm]

Moonshot announced they will be making weights available by July 27

https://twitter.com/Kimi_Moonshot/status/2077830229968683203

[efficax]

cost has nothing to do with why deepseek was disruptive, the fact that it means there is zero moat around anthropic or openai is what's disruptive about it. it means in the mid-term LLMs will be commoditized and customers will flock to the cheapest inference wherever they can find it. there's no reason to stick to the "frontier" labs

[rsanek]

> cost has nothing to do with it

> customers will flock to the cheapest inference

[efficax]

if deepseek cost twice as much to train it would prove the same thing: the american companies have no monopoly on state of the art llms, and commoditization is happening

[rsanek]

If AA is to be believed then per-task it is about the same cost as Sol. Agree that it's very different from DeepSeek v4 Pro, which is ~15x cheaper than K3.

https://artificialanalysis.ai/models/kimi-k3#price-cost

[hedora]

DeepSeek didn’t really change any trends though, unless you count the stock market.

It was impressive work, but models were commoditizing and inference costs were dropping rapidly already. They were neither the first nor the last 10x optimization, from what I’ve seen.

[stavros]

If you know of any other 10x optimisations currently, please let me know! I'm in the market for a model that's a tenth the price of a frontier model at the same level of quality.

[ac29]

You do understand that the "frontier" people are usually talking about is the cost-intelligence frontier right?

By definition there is no model that is both cheaper and as intelligent or better than another on the frontier.

[stavros]

OK, let me be more precise: If you know of a frontier model that's ten times cheaper than the previous frontier model at that level of intelligence, please let me know, I'm in the market for one.

Is that better?

[ac29]

Its just a single benchmark, but Luna 5.6 xhigh scores within the margin of error the same as Opus 4.8 max on DeepSWE for 8x cheaper. Luna max is quite a bit higher than Opus and still 4x cheaper

[bogdan]

To be fair the stock market is a big one

[fastball]

In my experience, the Chinese models are much more benchmaxxed than their frontier lab competitors, so I'm taking these results with a fairly large helping of salt.

[deanc]

That’s an interesting way to say you’re third. I’m only second to the ten other runners on my local Strava segments.

[simonw]

> In our evaluations, Kimi K3 delivers frontier-level performance

What page does that come from? I'm having trouble tracking it down.

[wolttam]

It was on the page linked in the top comment, but it's been removed.

[akoumjian]

Where are you seeing this write up?

[ekojs]

I copied that from https://platform.kimi.ai/docs/guide/kimi-k3-quickstart but it seems they updated the page to remove the benchmark score now.

[z3t4]

What hardware are they using to run the benchmarks?

[andai]

Where is this from?