@renatomoraesp@RyanGreenblatt@tszzl Because chinese labs were not at frontier level so far, they all released models months behind top world models. Their operating workflow is based on distilling GPT and Claude so until they change this workflow, they can't be a truly frontier lab/model.
@RyanGreenblatt My man - Opus 4.8 was launched on May 28th, 2026. Are you using the Mayan calendar? How do you conclude a model launched 2 months later, and in your own words is more capable, be 8 months behind?
Impressive mental gymnastics.
@RyanGreenblatt I'll be honest, k3 is above Opus by a big margin, idk how you are making this "predictions" of timelines, but I'm pretty sure that Kimi is getting really close to SOTA, I really believe they are around 2/3 months behind
Positive users praise Kimi K3 for surpassing Opus 4.8 by a wide margin, while negative users attack the analysis as uninformed cope and flawed on model timelines.
Based on 8 visible X reactions from 48 accounts; directional sample.
@renatomoraesp@RyanGreenblatt@tszzl Because chinese labs were not at frontier level so far, they all released models months behind top world models. Their operating workflow is based on distilling GPT and Claude so until they change this workflow, they can't be a truly frontier lab/model.
@RyanGreenblatt My man - Opus 4.8 was launched on May 28th, 2026. Are you using the Mayan calendar? How do you conclude a model launched 2 months later, and in your own words is more capable, be 8 months behind?
Impressive mental gymnastics.
@RyanGreenblatt I'll be honest, k3 is above Opus by a big margin, idk how you are making this "predictions" of timelines, but I'm pretty sure that Kimi is getting really close to SOTA, I really believe they are around 2/3 months behind
Kimi K3 was significantly but not massively above my expectations. I'd tentatively guess it's similar in overall usefulness/usability to Opus 4.8 and in overall capability somewhat above Opus 4.8 (while also being somewhat more benchmaxxed). As a pretrain, it's probably somewhere between 4.8 and Mythos (around halfway between?). Maybe this implies Kimi is like 8 or so months behind Anthropic in overall model strength/goodness (including usability) and like 6 or so months behind on overall capability (somewhat below Mythos Preview).
This gap is presumably reduced by distillation (and more generally using OpenAI/Anthropic models) and algorithm leakage/diffusion, so I think that hypothetically if the US completely stopped and recent algos didn't diffuse, it would maybe take Kimi like 10 months to fully catch up to the best internal (including in development) Anthropic model. (I think this notion might be a better measure of where Anthropic/OpenAI are relative to Kimi, even though this hypothetical won't happen.) And if the US completely stopped, it might take Kimi around 27 months to reach the level the US would otherwise have reached one year from now (as in, with a year of further progress).
My views here are pretty sensitive to how much benchmark performance is representative to overall usability.
I think I now expect an open-weight AI which is straightforwardly "Mythos-level at cyber" (including usability etc.) in like 5 months supposing Kimi and others don't change their open-weight model policy. (I don't have a strong view about how big of a deal this is for cyber, but it may cause significant political consequences. This could be a significant overestimate of the time required.)
I wonder what's driving Kimi being closer than I would have expected. Options include:
- Experiment compute is significantly less important than labor (and labor at Kimi is competitive, which seems super plausible)
- Implies more of a speedup from AI automating AI R&D and a bigger software-only intelligence explosion.
- Or possibly Kimi is just doing much better than US companies and this is overcoming experiment compute disadvantages.
- Algorithms are diffusing a lot / quickly (from e.g. OpenAI to Kimi).
- Perf is overstated / benchmaxxed a lot.
- Distillation / using OpenAI or Anthropic frontier AIs in AI development is very helpful for catching up. (But I'd guess Kimi K3 is a competitive pretrain which distillation doesn't help with?)
- US companies aren't going as fast as they could for whatever reason.
Kimi K3 was significantly but not massively above my expectations. I'd tentatively guess it's similar in overall usefulness/usability to Opus 4.8 and in overall capability somewhat above Opus 4.8 (while also being somewhat more benchmaxxed). As a pretrain, it's probably somewhere between 4.8 and Mythos (around halfway between?). Maybe this implies Kimi is like 8 or so months behind Anthropic in overall model strength/goodness (including usability) and like 6 or so months behind on overall capability (somewhat below Mythos Preview).
This gap is presumably reduced by distillation (and more generally using OpenAI/Anthropic models) and algorithm leakage/diffusion, so I think that hypothetically if the US completely stopped and recent algos didn't diffuse, it would maybe take Kimi like 10 months to fully catch up to the best internal (including in development) Anthropic model. (I think this notion might be a better measure of where Anthropic/OpenAI are relative to Kimi, even though this hypothetical won't happen.) And if the US completely stopped, it might take Kimi around 27 months to reach the level the US would otherwise have reached one year from now (as in, with a year of further progress).
My views here are pretty sensitive to how much benchmark performance is representative to overall usability.
I think I now expect an open-weight AI which is straightforwardly "Mythos-level at cyber" (including usability etc.) in like 5 months supposing Kimi and others don't change their open-weight model policy. (I don't have a strong view about how big of a deal this is for cyber, but it may cause significant political consequences. This could be a significant overestimate of the time required.)
I wonder what's driving Kimi being closer than I would have expected. Options include:
- Experiment compute is significantly less important than labor (and labor at Kimi is competitive, which seems super plausible)
- Implies more of a speedup from AI automating AI R&D and a bigger software-only intelligence explosion.
- Or possibly Kimi is just doing much better than US companies and this is overcoming experiment compute disadvantages.
- Algorithms are diffusing a lot / quickly (from e.g. OpenAI to Kimi).
- Perf is overstated / benchmaxxed a lot.
- Distillation / using OpenAI or Anthropic frontier AIs in AI development is very helpful for catching up. (But I'd guess Kimi K3 is a competitive pretrain which distillation doesn't help with?)
- US companies aren't going as fast as they could for whatever reason.
I did some quick tests that indicated that the Kimi K3 pretrain is around halfway between Opus 4 and Opus 4.5. So ~10 months behind Anthropic. These tests probably understate data improvements, so overall I think it's a similarly good pretrain to Opus 4.5 (~8 months behind). These tests are better at measuring "general pretrain capability" than at incorporating (coding-specific) data quality.
This was prompted by me thinking more and realizing that my claim that "As a pretrain, it's probably somewhere between 4.8 and Mythos (around halfway between?)" was probably too bullish on the model and that I might as well test and find out. (And yep, this was very wrong.)
I think Mythos is a pretty big step up in pretraining, so K3 might be more than 8 months behind on the historical pretraining trend relative to Mythos (as in, Mythos is >>3 months ahead of K3 and Mythos was fully done training ~5 months ago).
Overall, this makes me suspect more of the improvements are due to distillation-type effects and makes me think the full catch-up times would be somewhat longer (if Ant/OpenAI stopped but investment still followed current trends). Minimally, more of the improvement probably lives in post-training/mid-training.
For reference, the same test indicates K2.6 is around halfway between Sonnet 4 and Sonnet 4.5. (And this roughly corresponds to some other similar measures.)
Sorry about the error.
> Kimi K3 pretrain is around halfway between Opus 4 and Opus 4.5. So ~10 months behind Anthropic.
That is very likely true for irreducible compute cost
Anthropic's "pretraining progress" is first and foremost about scale
To clarify when I say "Kimi is like 8 or so months behind Anthropic in overall model strength/goodness", I'm trying to smooth the trajectories. Because Mythos Preview is way better than the prior AI Opus 4.6 (and probably much better than K3 in usefulness) and was ~5 months ago.
E.g., I think Opus 4.8 is maybe around 8 months or so behind Anthropic. Or possibly more like 6 or 7. But certainly >5! (And it will depend on how you do the accounting.)
Kimi K3 was significantly but not massively above my expectations. I'd tentatively guess it's similar in overall usefulness/usability to Opus 4.8 and in overall capability somewhat above Opus 4.8 (while also being somewhat more benchmaxxed). As a pretrain, it's probably somewhere between 4.8 and Mythos (around halfway between?). Maybe this implies Kimi is like 8 or so months behind Anthropic in overall model strength/goodness (including usability) and like 6 or so months behind on overall capability (somewhat below Mythos Preview).
This gap is presumably reduced by distillation (and more generally using OpenAI/Anthropic models) and algorithm leakage/diffusion, so I think that hypothetically if the US completely stopped and recent algos didn't diffuse, it would maybe take Kimi like 10 months to fully catch up to the best internal (including in development) Anthropic model. (I think this notion might be a better measure of where Anthropic/OpenAI are relative to Kimi, even though this hypothetical won't happen.) And if the US completely stopped, it might take Kimi around 27 months to reach the level the US would otherwise have reached one year from now (as in, with a year of further progress).
My views here are pretty sensitive to how much benchmark performance is representative to overall usability.
I think I now expect an open-weight AI which is straightforwardly "Mythos-level at cyber" (including usability etc.) in like 5 months supposing Kimi and others don't change their open-weight model policy. (I don't have a strong view about how big of a deal this is for cyber, but it may cause significant political consequences. This could be a significant overestimate of the time required.)
I wonder what's driving Kimi being closer than I would have expected. Options include:
- Experiment compute is significantly less important than labor (and labor at Kimi is competitive, which seems super plausible)
- Implies more of a speedup from AI automating AI R&D and a bigger software-only intelligence explosion.
- Or possibly Kimi is just doing much better than US companies and this is overcoming experiment compute disadvantages.
- Algorithms are diffusing a lot / quickly (from e.g. OpenAI to Kimi).
- Perf is overstated / benchmaxxed a lot.
- Distillation / using OpenAI or Anthropic frontier AIs in AI development is very helpful for catching up. (But I'd guess Kimi K3 is a competitive pretrain which distillation doesn't help with?)
- US companies aren't going as fast as they could for whatever reason.
I did some quick tests that indicated that the Kimi K3 pretrain is around halfway between Opus 4 and Opus 4.5. So ~10 months behind Anthropic. These tests probably understate data improvements, so overall I think it's a similarly good pretrain to Opus 4.5 (~8 months behind). These tests are better at measuring "general pretrain capability" than at incorporating (coding-specific) data quality.
This was prompted by me thinking more and realizing that my claim that "As a pretrain, it's probably somewhere between 4.8 and Mythos (around halfway between?)" was probably too bullish on the model and that I might as well test and find out. (And yep, this was very wrong.)
I think Mythos is a pretty big step up in pretraining, so K3 might be more than 8 months behind on the historical pretraining trend relative to Mythos (as in, Mythos is >>3 months ahead of K3 and Mythos was fully done training ~5 months ago).
Overall, this makes me suspect more of the improvements are due to distillation-type effects and makes me think the full catch-up times would be somewhat longer (if Ant/OpenAI stopped but investment still followed current trends). Minimally, more of the improvement probably lives in post-training/mid-training.
For reference, the same test indicates K2.6 is around halfway between Sonnet 4 and Sonnet 4.5. (And this roughly corresponds to some other similar measures.)
Sorry about the error.
MoonshotAI will overtake OpenAI and Anthropic before the end of the year
or will they? at least that's what the hype kiddies on X want you to believe
So let me make it falsifiable. They are saying:
- China / MoonshotAI is catching up
- they are catching up generally (not just coding, but almost all domains and including restricted models like Mythos 5)
- the gap is currently ~1.4 months based on Artificial Analysis Index and benchmarks provided by MoonshotAI, where Kimi-K3 beats Opus 4.8 in 30 of 35 benchmarks, and GPT-5.6-Sol in 19 of 35 benchmarks
(they ignore the existence of all Mythos variants)
- China is not catching up due to distillation, so they should overtake US labs
Their implicit prediction then is:
- a chinese model / MoonshotAI will overtake Anthropic (and OpenAI) on the Artificial Analysis Index by:
- Median: 2026-12-24 (80% CI: 2026-09-17, 2027-09-14)
Since they claim that chinese models are as general as american models, we should see unsolved mathematics, physics, and more being solved by chinese models at higher rates than american models.
Speaking to its generality Kimi-K3 should surpass Opus 4.8 and GPT-5.6-Sol on the majority of these benchmarks:
- METR Time Horizons, FrontierCode, MirrorCode, UK AISI cyber ranges, ExploitBench/ExploitGym, CritPT, FrontierMath T4, ARC-AGI-2 / ARC-AGI-3, WeirdML, ALE-Bench, GSO, MRCR2/GraphWalks
- vibes
---
Some other things that are more speculative and downstream of China overtaking US models:
- more involvement by the USG
- stricter export controls on semis
- potentially a Manhatten-style project, as we will be behind in 2027 and are racing against China
- also in the cards: US banning chinese models or US labs distilling from chinese models
---
I have already stated my position clearly.
Chinese models are generally ~6-8 months behind, with some domains like coding behind slightly less.
Kimi-K3 did not significantly shift my estimate on the gap and it currently does not change my outlook on the future, but we will have a MUCH clearer picture once we have all the benchmarks I mentioned earlier.
The main reasons for my position:
- Kimi-K3 doesn't even beat Mythos Preview, a ~5 month old model
- We will likely not see much larger open models than Kimi-K3 for several months, likely not until early-mid 2027
- Meanwhile Anthropic is sitting on a 10T model since ~February, OpenAI likely just finished the training of GPT-6, which should also be around that size, and more 10T param US models are coming from SpaceX AI, Google and Meta.
- We are currently not seeing the true frontier of models. Anthropic and OpenAI are currently sandbagging as the legal situation for releasing new frontier models is unclear.
- Historically, chinese models have been more benchmaxxed than US models, meaning their benchmark numbers do not translate to real world performance as well as their american counterparts
- GPT-5.6-Sol is still 2-3x more token-efficient on the Artificial Analysis Index than Kimi-K3 (while likely being smaller, ~2T)
- US labs have more compute
---
I'm very happy that Kimi bros released this model.
It's a great model and probably the first really useful chinese model.
To clarify when I say "Kimi is like 8 or so months behind Anthropic in overall model strength/goodness", I'm trying to smooth the trajectories. Because Mythos Preview is way better than the prior AI Opus 4.6 (and probably much better than K3 in usefulness) and was ~5 months ago.
Kimi K3 was significantly but not massively above my expectations. I'd tentatively guess it's similar in overall usefulness/usability to Opus 4.8 and in overall capability somewhat above Opus 4.8 (while also being somewhat more benchmaxxed). As a pretrain, it's probably somewhere between 4.8 and Mythos (around halfway between?). Maybe this implies Kimi is like 8 or so months behind Anthropic in overall model strength/goodness (including usability) and like 6 or so months behind on overall capability (somewhat below Mythos Preview).
This gap is presumably reduced by distillation (and more generally using OpenAI/Anthropic models) and algorithm leakage/diffusion, so I think that hypothetically if the US completely stopped and recent algos didn't diffuse, it would maybe take Kimi like 10 months to fully catch up to the best internal (including in development) Anthropic model. (I think this notion might be a better measure of where Anthropic/OpenAI are relative to Kimi, even though this hypothetical won't happen.) And if the US completely stopped, it might take Kimi around 27 months to reach the level the US would otherwise have reached one year from now (as in, with a year of further progress).
My views here are pretty sensitive to how much benchmark performance is representative to overall usability.
I think I now expect an open-weight AI which is straightforwardly "Mythos-level at cyber" (including usability etc.) in like 5 months supposing Kimi and others don't change their open-weight model policy. (I don't have a strong view about how big of a deal this is for cyber, but it may cause significant political consequences. This could be a significant overestimate of the time required.)
I wonder what's driving Kimi being closer than I would have expected. Options include:
- Experiment compute is significantly less important than labor (and labor at Kimi is competitive, which seems super plausible)
- Implies more of a speedup from AI automating AI R&D and a bigger software-only intelligence explosion.
- Or possibly Kimi is just doing much better than US companies and this is overcoming experiment compute disadvantages.
- Algorithms are diffusing a lot / quickly (from e.g. OpenAI to Kimi).
- Perf is overstated / benchmaxxed a lot.
- Distillation / using OpenAI or Anthropic frontier AIs in AI development is very helpful for catching up. (But I'd guess Kimi K3 is a competitive pretrain which distillation doesn't help with?)
- US companies aren't going as fast as they could for whatever reason.
57 comments
245 saves
47 reposts
Sentiment
Positive25%75%Negative
Positive Read
Positive users praise Kimi K3 for surpassing Opus 4.8 by a wide margin, while negative users attack the analysis as uninformed cope and flawed on model timelines.
Based on 8 visible X reactions from 48 accounts; directional sample.
I did some quick tests that indicated that the Kimi K3 pretrain is around halfway between Opus 4 and Opus 4.5. So ~10 months behind Anthropic. These tests probably understate data improvements, so overall I think it's a similarly good pretrain to Opus 4.5 (~8 months behind). These tests are better at measuring "general pretrain capability" than at incorporating (coding-specific) data quality.
This was prompted by me thinking more and realizing that my claim that "As a pretrain, it's probably somewhere between 4.8 and Mythos (around halfway between?)" was probably too bullish on the model and that I might as well test and find out. (And yep, this was very wrong.)
I think Mythos is a pretty big step up in pretraining, so K3 might be more than 8 months behind on the historical pretraining trend relative to Mythos (as in, Mythos is >>3 months ahead of K3 and Mythos was fully done training ~5 months ago).
Overall, this makes me suspect more of the improvements are due to distillation-type effects and makes me think the full catch-up times would be somewhat longer (if Ant/OpenAI stopped but investment still followed current trends). Minimally, more of the improvement probably lives in post-training/mid-training.
For reference, the same test indicates K2.6 is around halfway between Sonnet 4 and Sonnet 4.5. (And this roughly corresponds to some other similar measures.)
Sorry about the error.
> Kimi K3 pretrain is around halfway between Opus 4 and Opus 4.5. So ~10 months behind Anthropic.
That is very likely true for irreducible compute cost
Anthropic's "pretraining progress" is first and foremost about scale
To clarify when I say "Kimi is like 8 or so months behind Anthropic in overall model strength/goodness", I'm trying to smooth the trajectories. Because Mythos Preview is way better than the prior AI Opus 4.6 (and probably much better than K3 in usefulness) and was ~5 months ago.
E.g., I think Opus 4.8 is maybe around 8 months or so behind Anthropic. Or possibly more like 6 or 7. But certainly >5! (And it will depend on how you do the accounting.)
