Rendered at 05:48:50 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
Ariarule 10 hours ago [-]
Always glad to see more open-weight models, but this caption on the 2nd demo image had me do a double-take: "Land or Water Generalization Experiment: We recreated the viral X puzzle by asking Beam to create a fixed 180×90 grid for longitudes -179° to 179° and latitudes -89° to 89°, with 16,200 points. This puzzle is a few days old, so could not appear in the training data, thus testing the model’s generalization. Beam gets 95.5% coverage right, putting us between Opus 5 (92.5%) and Fable 5 (97.8%), which shows how well it generalizes to novel new tasks."
Oof, no, this "puzzle is a few days old" is incorrect even if it's a social media trend just recently. Asking a model to generate a world map in this way is _at least_ from August 2025 as it appeared on LessWrong at that time: https://www.lesswrong.com/posts/xwdRzJxyqFqgXTWbH/how-does-a...
extr 10 hours ago [-]
Yeah I remember when the original post about this came out. Def not recent. Though I think their point survives in that they didn't exactly RL on this.
glitchc 4 hours ago [-]
It could be old but still not be part of the training set.
jiggawatts 8 hours ago [-]
I'd love to see these tests repeated for the current frontier models...
charlieyu1 9 hours ago [-]
I don't think age of the puzzle even matters, all models have search capacities these days
criemen 9 hours ago [-]
> all models have search capacities these days
one would hope that they disable websearch and internet access (maybe all tools?) when doing generalization testing?
zakisaad 7 hours ago [-]
Model weights (what is being tested here) don't inherently "access the web" when inference is running. If the model has access to a web search tool, that's a different story.
dexwiz 9 hours ago [-]
Is search part of the model or the harness?
Cycl0ps 8 hours ago [-]
Maybe that's a rhetorical question but just in case - the search would always be part of the harness. A model is only handling next-token prediction for a given input. That token may be something like [[web search]] to invoke a tool call but the actual call would be handled by the harness.
dexwiz 4 hours ago [-]
Yeah it was rhetorical. Search would be implemented as a tool call. Pure intelligence tests would likely have limited tools. But maybe they would have a python sandbox to solve issues like Rs in strawberry.
6 hours ago [-]
Barbing 3 hours ago [-]
If they made that statement and knowingly had search enabled, it would essentially be fraudulent.
htrp 10 hours ago [-]
> Beam is a sparse Mixture-of-Experts model with 501 billion total parameters, 23 billion active, built for coding, reasoning, and agentic workloads.
> Beam’s capabilities come from major investments in both pretraining and reinforcement learning (RL). We pretrained the model on 23.8 trillion diverse, curated, high-quality tokens from the web and proprietary licensed datasets, matching or outperforming available similar-sized open base models. In parallel, we developed the algorithms, training environments, and infrastructure needed to sustain high-compute RL at exceptional scale. Our high-compute RL run generated over 100 million rollouts on 10.5K NVIDIA GB300 GPUs over 4 weeks of training.
Early access, no weights no tech details, just a sign up here for info
wronglebowski 10 hours ago [-]
I'm all for more open models, but talk is cheap and this is a rather pointless announcement without anything backing it up. Publish your weights and HF repo or shut up IMO.
antoniojtorres 6 hours ago [-]
Is that all that a company about to give away the product of 10,000 GPUs running for a month gets to be now? give it away without a single promotional post or shut up? I support open source as much as the person but this is pretty caustic.
mewse-hn 2 hours ago [-]
I am also starting to take issue with "we've developed a new cutting edge model, and nobody can use it" announcements. Most recently with Google's Argon, at least they were using it internally and they'd be slowly rolling it out. This one is from a company I've never heard of, they're not releasing weights or offering API access at this point, it feels like a fairly worthless announcement.
nl 1 hours ago [-]
There have been previous "we will release the model" when the model release never comes in the past.
I think the comment you are replying to is unnecessarily hostile too though.
5 hours ago [-]
3 hours ago [-]
zelphirkalt 10 hours ago [-]
And also a "proprietary data set" hahaha... Probably just means they don't want to show it, and it is data, that either they shouldn't have, or that there is nothing special about their training data and it is just meant to sound like there is some secret ingredient, while there is none.
nl 1 hours ago [-]
This isn't true.
Companies pay lots of money for proprietary agentic trajectories which are used during RL. These are things like "Task: summarize stock levels for months end accounting" which then traces the task though using SAP to look at different SKU stock levels, exporting them and generating summary Excel spreadsheets.
This is very different to the "scrape the internet" datasets that a table stakes for training a LLM.
Not sharing the data is pretty standard because 1) it tends to get the lawyers involved and 2) good data is critical for getting good results.
Imo you can get better results with great data and generic modeling techniques than with incredible modeling techniques and crappy data. Because if you have crappy data, you won’t even know if your model is good because your evals will also be bad.
This is why Anthropic is throwing a fit about the Chinese distillation “attacks”. Clean reasoning traces are gold.
mlmonkey 8 hours ago [-]
Data has copyright issues, so one can't share it generally without getting permissions from all of the copyright holders. The data is not theirs to share, anyways. The derived (learned) weights are a different matter.
ordersofmag 3 hours ago [-]
True for the pre-training data scraped from diverse sources. Less so for the later stage data for RL which is by all account more of a differentiator. In most cases the labs themselves produced the data so they are the copyright holders (or they are borrowing it from other labs via distillation). A lot of it is synthetic data, and since you can't copyright AI output, it becomes less about copyright and more about trade secrets.
vanuatu 9 hours ago [-]
this is very normal for frontier lab companies. you need good data either synthetic or labelled (all the chinese open source models have their own armies of data labelers)
Loquebantur 9 hours ago [-]
> Beam is undergoing final red-teaming and evaluations. You can sign up here for early access to the model.
> We will release the weights, technical report, model card, and developer artifacts later this month.
wren6991 7 hours ago [-]
I thought it would be interesting to look at some key figures vs another contemporary model in the same weight class (DeepSeek V4.1 Flash)
DS V4.1F Beam
LM total params 552B 501B
LM active params (prefill) 8B 23B
LM active params (decode) 16B 23B
N-gram/PLE params 196B 0
Pretrain tokens 45T 28T
Disk KV bytes/token (FP4) 890 No information
Vision Yes (pretrain) No
Weights available Yes (launch day) "This month"
Weights licence MIT Apache 2.0
At first blush the benchmarks are impressive, but to paraphrase Linus: "Talk is cheap, show me the weights." :-)
laybak 7 hours ago [-]
yeah agreed, let's normalize "show me the weights" in this era
NorwegianDude 10 hours ago [-]
Bigger and still worse than existing free Chinese models that are smaller? Open weight models are nice, but at this point it seems western models are very far behind Chinese ones, despite Chinese companies publishing a lot of their findings. I hope we get more open models and more providers, as being stuck with a model from China or US with no competition is risky.
Google does do a great job with Gemma models. It's one of the few language models actually good at language. OpenAI's top closed models can't even write norwegian correctly.
mirekrusin 9 hours ago [-]
It takes time / few iterations to get it right (and it's moving target), but yes, expensive trial, my personal feeling is that they went a bit too high, at the same time who knows, maybe good move – as they're saying RL didn't plateau. It feels like they had something like $100M budget for it?
Iolaum 22 minutes ago [-]
I am really curious why after Qwen-3.8-flash and deepseek-v4.1-flash newer open weight models (or proprietary but they don't tell) don't use n-grams. It looks (to me) like they are a cheep way to add more knowledge to the model.
What do I mean by cheap? You can rely on the SSD to retrieve the relevant tokens as no computation is needed meaning you can leverage storage (or cpu ram if you don't have unified memory) to serve part of the model which (to my understanding) is much cheaper to get than GPU RAM.
Anyone know what am I missing? Or is it that the pace of iteration for labs slow enough that they can't actually leverage it yet?
onlyrealcuzzo 10 hours ago [-]
This appears to be larger than DeepSeek v4.1 Flash, more expensive to run, and worse on every measured metric.
Am I missing something?
swiftcoder 9 hours ago [-]
> Am I missing something?
It's pretty clear from their framing ("Beam advances the Western open-weight frontier") that one of their main selling points is not being a Chinese lab.
I can't imagine that mattering to many individuals, but I guess someone out there has a government contract that forbids the use of foreign models
htrp 9 hours ago [-]
Reflection raised on the idea of creating the "American Deepseek Project"
I still don't get why, after over a decade on HN, people refuse to Google very simple questions.
jauntywundrkind 8 hours ago [-]
Reciprocally, people probably do go find out. The great filter is also how many of them come back to post the useful interesting information.
hadlock 5 hours ago [-]
I still don't get why, after 25 years of slashdot, still ask why people DRTFA and complain about it as meta commentary.
InB4: kids these days :shakes-fist-at-cloud:
nickpsecurity 2 hours ago [-]
Sometimes we just enjiy having a conversation with people. It's how we got many answers before Google exists. I believe such choices have many, positive effects on people that society is losing.
NamlchakKhandro 6 hours ago [-]
because my templeos goes straight from ring0 after bios straight into a ui for hackernews that only lets me scroll, click into comments and type comments.
JSR_FDED 4 hours ago [-]
Try using the same Internet that you used to download templeos to search for stuff.
dominotw 8 hours ago [-]
quesiton is more like "lets analyze the motivations behind funding this"
kelnos 6 hours ago [-]
And a much better post would be "Just checked, and X, Y, and Z are funding this. My bet is that Y is funding it because $REASON..."
Asking an easily-searchable question is just lazy.
calmworm 2 hours ago [-]
Profit. Always.
nozzlegear 6 hours ago [-]
I don't get why anyone posts anything on HN when they can just have an LLM generate an entirely self-contained conversation now.
/s
The conversation is why.
mirekrusin 9 hours ago [-]
Multiple independent approaches are cool and all but fully open source model training (datasets, pipeline, checkpoints) should be taking advantage of being open and share runs/budget between different entities.
dotancohen 10 hours ago [-]
We're still at the stage where every new entrant is welcome in my opinion. Doesn't need to be record-breaking upon initial release.
flockonus 9 hours ago [-]
It depends! If a startup is entering with a large model to face other larger models, it must be better at least in 1 meaningful dimension.
500B params performing worse than other OSS of the same size is pretty meaningless if no one will use it.
halJordan 9 hours ago [-]
I disagree. Sure let them play and see if they can improve. But this model has more compute and more training data than the predecessors it fails to surpass. That only means their training regime is inferior if their predecessors did so much more with so much less. That inferiority should not be encouraged.
kelnos 6 hours ago [-]
You don't just magically do better than everyone else on every metric on your first go at something. Doing worse than others and refining is how pretty much everything works.
halJordan 4 hours ago [-]
I think i captured that in my first (second?) sentence
janalsncm 9 hours ago [-]
The reality is they trained a model and it looks worse on benchmarks than Qwen or GLM. I don’t see how sharing the weights hurts anyone? Even when Llama 4 came out and it was a dumpster fire, it didn’t affect me personally.
> That only means their training regime is inferior if their predecessors did so much more with so much less
Hard to imagine how that wouldn’t be the case. They probably missed the boat on distilling Claude (or their lawyers said no), they probably didn’t hire an army of math PhDs to write reasoning traces, they don’t have millions of DAUs in a coding agent to train from, and they probably have less money, less experience, fewer top tier researchers, and fewer resources for experiments. They are an underdog without a doubt.
None of that means they shouldn’t release their model.
halJordan 4 hours ago [-]
Them releasing the weights doesn't hurt anyone. It's the peanut gallery clamoring to put them onto the same pedestal as actual tier 1 companies simply because they aren't named openai or anthropic that is hurtful.
thinkcontext 7 hours ago [-]
I wonder if that's an indication that they are not distilling which limits how good they can get.
halJordan 4 hours ago [-]
Openai, grok, and Anthropic aren't distilling. Theyre just second class. It's not a big deal, we just shouldn't be lauding them for being second class.
girvo 21 minutes ago [-]
> Openai, grok, and Anthropic aren't distilling
Says who? We know Grok does at the least. They admitted it openly.
thinkcontext 3 hours ago [-]
Musk said under oath that they use distillation for Grok.
vanuatu 9 hours ago [-]
Reflection is explicitly marketed as the 'US' DeepSeek
seems like they are aiming to provide both inference and RLaaS for american companies and western govts. even if they never fully beat deepseek if they get close enough the fact that they're American will help them close deals
aizk 9 hours ago [-]
Yes it's (hopefully) not distilled from every single major American provider.
jstummbillig 10 hours ago [-]
Apparently there is more to making good models than copying everything on the internet.
chews 4 hours ago [-]
12 yards long, 2 lanes wide,
65 tons of American Pride!
Canyonero! Canyonero!
Centigonal 9 hours ago [-]
new entrant in this weight class, US lab.
martini333 10 hours ago [-]
Beam goes brrrr
algoth1 3 hours ago [-]
Does this one also routes to Claude under the hood like Reflection 70B did? I recall they even run a basic regex to remove "Claude" from the output. Then they promised to be completely transparent on the postmortem (they claim they had no idea what had happened), but the postmortem never came
eldenring 2 hours ago [-]
Reflection AI is a completely different company with no relation (as far as I can tell) to the model you are referencing from 2024
drubs 10 hours ago [-]
I remember being in the room with pretraining day 1 to help monitor the training job launch. Watching this model train from day 1 has been an amazing experience!
jeremyjh 9 hours ago [-]
What sort of outputs or telemetry is monitored on a large pre-training run?
drubs 9 hours ago [-]
Outside of ML metrics, you're monitoring the health of every piece of hardware in the system. You need to make sure that you have every GPU, every CPU, the PCIe buses, the networking fabric are all working without any errors. You need to ensure that you can respond as fast as possible to any possible error. One bad component can bottleneck the entire job.
I really enjoyed reading the log book from the training of OPT-175B at Meta… I guess it’s all classified info but it’d be fun to read a blog post about the crazy day to day issues you run into when doing things at this scale :)
spindump8930 4 hours ago [-]
GPU failures are frequent enough that at a certain scale, you constantly have workers dropping out. Designing systems that can still keep training is very interesting!
efficient_dairy 2 hours ago [-]
I don't find open-weight models that impressive anymore. MiMo-V2.6 already showed that you can have a not so crazy architecture and enough compute, the bottleneck is then just the data. OAI and Anthropic are largely the frontier models because of the synthetic data they made. They have a large customer base and have the user's data as well as knowing what tasks their customers use the models for and what domain they should get synthetic data for.
computerdork 2 hours ago [-]
Interesting. Have heard about the need to create synthetic data for LLM's, but didn't realize it was such an important factor. Although, how big a factor synthetic data is the next question, but guess there is no way to truly verify how much difference it makes with these closed models.
efficient_dairy 2 hours ago [-]
This is literally what the labs have been doing over the past year. The whole idea of emergence is a lie, there is some interpolation and superhuman long evaluations the models can do, but almost all the gains are from synthetic data. They hire thousands of professionals and pay them as much as 200$/hour to create many tasks that they want the model to perform and use these to teach the model on how to do it with RL. OAI had 30,000 contractors from Mercor for Sol 5.6.
When you see Opus suddenly becoming great at blender or some other 3D graphics, that's because they hired professionals and had them do similar tasks that people are looking for. They keep having better professionals at each iteration and so the quality improves. There is no emergence or "General" intelligence. The model doesn't learn to become better at a task because of scaling laws or emergence or whatever they might wanna say, it is literally RL on tasks that they want the model to perform well on.
computerdork 46 minutes ago [-]
Also interesting. Do you happen to know if once they've used those thousands of contractors to teach the model something like Blender (or some similar app) using RL, when they training a new model, do they need to use same contractors again to teach that new model the same behaviors?
The reason I ask is even though it can take hundreds or thousands of contractors to teach a model a certain behavior, wonder if they really only need to do it once for each desired behavior? (of course, future models might expand and refine this previous training) Because if that is the case, then wow, then future models can really expand their capabilities really very fast.
...Am wondering if they somehow record a training session so they can play it back whenever they need to train a new model with the same info? Or maybe the new models can just use distillation from the old model to relearn the old behaviors?
efficient_dairy 6 minutes ago [-]
The contractors don't directly teach the model. They create datasets. Mostly they create tasks within an environment that the current generation of models wouldn't be able to do, they then write maybe a solution, a set of rules for evaluating the response, and whatever is needed for the RL. These tasks form a dataset. You can see examples of a task in the Mimo dataset that was open-sourced recently. The dataset would then be used by engineers for post-training of whatever model. Some of the model iterations that are released every month tend to be just a further post-training of the same base model that was pre-trained months ago. OAI recently has been doing a lot more pre-training, but for a long time they had the same pre-trained base model. This is why you see so many releases done so fast by the labs, they just need to post-train the same base model on whatever task they think would be better suited, I would speculate based on what users want and what the benchmarks test for.
Regarding your second question, I think if they want to further post-train a model using new data, they wouldn't feel the need to re-train it again on the data that it has already being trained on. But you never know. If the model is a completely new pre-trained base model, then they could either train the model using all the data and/or use a previous model to teach it. There is definitely a bunch of tricks they do to evaluate the models and check the performance or whatever their recipe is. It's really up to what the engineers would prefer. But you get the core idea, the models are not suddenly coming up with how to use the Blender on their own, they are explicitly being trained on a dataset curated by a professional that teaches the model how to use Blender. Surely there is another aspect that if the model gets better at coding, then it also helps it become better at Blender, and you have that transfer learning. However, there is no emergence or a deity popping up. But you get people who were evaluating theses models on blender use and suddenly seeing the model ace their tasks and they think they are dealing with a super-intelligence. They then undergo an AI psychosis once they try and extrapolate the (super)-exponential improvement in that one task over the next few months and across all other domains.
Regarding your third question, I already answered at it. But, when it comes to training, they definitely freeze the weights after each run just in case an issue arises and they need to address it (a GPU not working or the loss value blowing up).
shingoshoji 52 minutes ago [-]
I'm a big fan of open-weight models.
It's true that no benchmark communicates the whole picture, and we won't really know how it behaves until weights are out, but the performance here doesn't seem particularly groundbreaking just based on the benchmark.
michaelkdev 9 hours ago [-]
Is it worse than the top open-weight Chinese models? Yes, it is, but at least the West has joined the party, and hopefully they will iterate on this and keep up the pace. The Chinese labs will certainly release new and powerful versions soon, so it's all about relative pace right now.
reissbaker 3 hours ago [-]
Instead of yet another mediocre but fully-made-in-the-West open model (alongside Mistral, Trinity, Poolside, Inkling, etc etc) I'd really love for a Western neloab start the same way Qwen did: by focusing on post-training. Qwen's first release was a Llama 1 finetune [1]! Once they made it useful, they started working their way back in the stack to also do their own pretraining, etc. Starting with pretraining feels like such a waste: there's millions of dollars of crystallized compute and data sitting around in the Chinese model weights. Why not start with one of those, and only work your way back to pretraining once you've released something you can prove is useful?
I think there aren't any recent base models to post train on. Labs don't release them any more.
reissbaker 3 hours ago [-]
At least for RL, you don't need a base model — the rollouts are run in an inference engine with an instruction-tuned model using a chat template. You can start with just that!
redox99 52 minutes ago [-]
Yeah but they are already RL'd too. No doubt you can further improve an RL'd model by doing your own better RL on top of that. But it's not going to be the same as starting with a base model or instruction tuned model.
I am a great fan of open weights model, but off late I am starting to loose track on the capabilities of the latest models released and now a days most of the open weights models are above 100 B params which is not going to run in our laptops. What happened to Jev hype? A 501B Beam model is not going to help with it (Non AR Schema driven).
I am more interested with a SOTA Frontier 8B-10B model. Is this even possible?
astrostl 6 hours ago [-]
Beggar choosing: my kingdom for more 90B-133B MoE local models. Especially with disk offload, that is a function/performance sweet spot for Mac workstations with 64GB-128GB of RAM.
eaf7e281 8 hours ago [-]
> Where frontier open models like Kimi K3 remain ahead on raw capability, Beam's advantage is efficiency at inference time.
It's great to see a company that acknowledges it still needs improvement instead of making false claims.
arjunchint 1 hours ago [-]
ok so they are comparing themselves to and claim to be beating GLM's last generation GLM 5.2 model, GLM 5.3 Flash is a monster, this is honestly embarassing
4 hours ago [-]
aeetes 10 hours ago [-]
the performance chart puts the better open source models behind the fold making it seem like it outperforms them... but it doesn't! all for open source models but this announcement is misleading
brumbelow 10 hours ago [-]
Yes. All the link made me realize is that I should checkout Deepseek 4.1 flash
segmondy 8 hours ago [-]
Any time a new lab shows up, folks complain about how their models are worse. Really? It would be nice if a new comer comes from no where and beats everyone, but that's rarely the case. The good thing is that other labs/people are figuring out how to build this, and if they keep at it then this is as bad as it gets for them and it would hopefully get better. A new entrant to the market is good for everyone.
spindump8930 4 hours ago [-]
Reflection isn't really "from no where", they have huge financial backing, 10K of the latest GPUs, and many ex leads from the established labs.
NamlchakKhandro 6 hours ago [-]
honest question: how honest do you think people are about their improvements and performance compared to objective results when all you do is praise them?
participation awards are not helpful.
andai 3 hours ago [-]
Noticed some good questions made dead at the bottom of this thread. Odd...
hidelooktropic 7 hours ago [-]
Nitpicking but I really wish this benchmarks table were easier to read. Should show which columns win in each row and should not require horizontal scrolling to see across.
minimaxa 7 hours ago [-]
That was exciting... Will come back later when the weights are out and gguf'd.
Access is currently limited. We'll contact you if early access becomes available.
hypfer 9 hours ago [-]
Someone should name their next model "Workhorse" just for SEO reasons.
It's interesting how the industry converged to this very term, given that very less work is being done by horses since quite a while.
latentsea 4 hours ago [-]
I guess future AI agents might market things as a "workman" for the same reason, despite less work being done by people :)
hnedeotes 9 hours ago [-]
Thankfully there wasn't mistagging, we could have ended with workjackass.
zopper 10 hours ago [-]
Open model that is not yet open or widely accessible via API. Primarily comparing to non-SOTA models like Inkling and GLM 5.2. Included comparison to GLM 5.3 and DeepSeek V4.1 Flash in the table, but not in the charts (I assume they would make them look bad). Also no results from AA Index or Arena.
keeganpoppen 10 hours ago [-]
very curious to see more about what kinds of hardware you can run this on and the perf. characteristics… on the face of it, it seems like optimizing for inference speed might(?) be good for running on smaller hardware, but i suppose it could be the other way around and it is actually much resource-hungrier for the number of parameters, etc. …
TheArcane 8 hours ago [-]
If you don't buy into "America good, China bad" narrative, this new entrant & release by Inclusion Ai is a lot more exciting by every measurable metric.
There are a lot of open-research on pre-training, post-training and RL data mixtures and sourcing.
I recommend checking papers from Datalogy, Nvidia Nemotron, Ai2 (Ollmo, Tulu, ...) and the recent model from Aleph Alpha if you want to learn more.
spindump8930 4 hours ago [-]
All of those are good references. Other folks in the thread are missing distinctions between pre training (~the internet + curated sources) and post training (~instructions and RL)
wren6991 7 hours ago [-]
I heard you should ask Claude about this. Preferably with thousands of accounts, routed through residential proxies
altcognito 10 hours ago [-]
If you ask a model, they will generally tell you where to get data. Modern frontier models have the large advantage of having tens if not hundreds of millions of users providing use cases to train against to improve their responses.
You can also hire teams to create data for you for higher quality.
Hamuko 10 hours ago [-]
Get data from Claude. That's what the Chinese (allegedly) do.
kbwal7 10 hours ago [-]
Note that this sort of distillation is NOT for pre-training data (which is tens of trillions of tokens). I think the allegations against Chinese companies by Anthropic is more so that they distill SFT data (which is good for post-training, but you still need a strong base model)
Oof, no, this "puzzle is a few days old" is incorrect even if it's a social media trend just recently. Asking a model to generate a world map in this way is _at least_ from August 2025 as it appeared on LessWrong at that time: https://www.lesswrong.com/posts/xwdRzJxyqFqgXTWbH/how-does-a...
one would hope that they disable websearch and internet access (maybe all tools?) when doing generalization testing?
> Beam’s capabilities come from major investments in both pretraining and reinforcement learning (RL). We pretrained the model on 23.8 trillion diverse, curated, high-quality tokens from the web and proprietary licensed datasets, matching or outperforming available similar-sized open base models. In parallel, we developed the algorithms, training environments, and infrastructure needed to sustain high-compute RL at exceptional scale. Our high-compute RL run generated over 100 million rollouts on 10.5K NVIDIA GB300 GPUs over 4 weeks of training.
Early access, no weights no tech details, just a sign up here for info
I think the comment you are replying to is unnecessarily hostile too though.
Companies pay lots of money for proprietary agentic trajectories which are used during RL. These are things like "Task: summarize stock levels for months end accounting" which then traces the task though using SAP to look at different SKU stock levels, exporting them and generating summary Excel spreadsheets.
This is very different to the "scrape the internet" datasets that a table stakes for training a LLM.
Xiaomi released a fairly developer-centric dataset like this here: https://huggingface.co/datasets/XiaomiMiMo/MiMo-V2.6-RL-oss
SpreadsheetRL is another fairly specialized dataset: https://spreadsheet-rl.github.io/
Imo you can get better results with great data and generic modeling techniques than with incredible modeling techniques and crappy data. Because if you have crappy data, you won’t even know if your model is good because your evals will also be bad.
This is why Anthropic is throwing a fit about the Chinese distillation “attacks”. Clean reasoning traces are gold.
> We will release the weights, technical report, model card, and developer artifacts later this month.
Google does do a great job with Gemma models. It's one of the few language models actually good at language. OpenAI's top closed models can't even write norwegian correctly.
What do I mean by cheap? You can rely on the SSD to retrieve the relevant tokens as no computation is needed meaning you can leverage storage (or cpu ram if you don't have unified memory) to serve part of the model which (to my understanding) is much cheaper to get than GPU RAM.
Anyone know what am I missing? Or is it that the pace of iteration for labs slow enough that they can't actually leverage it yet?
Am I missing something?
It's pretty clear from their framing ("Beam advances the Western open-weight frontier") that one of their main selling points is not being a Chinese lab.
I can't imagine that mattering to many individuals, but I guess someone out there has a government contract that forbids the use of foreign models
InB4: kids these days :shakes-fist-at-cloud:
Asking an easily-searchable question is just lazy.
/s
The conversation is why.
500B params performing worse than other OSS of the same size is pretty meaningless if no one will use it.
> That only means their training regime is inferior if their predecessors did so much more with so much less
Hard to imagine how that wouldn’t be the case. They probably missed the boat on distilling Claude (or their lawyers said no), they probably didn’t hire an army of math PhDs to write reasoning traces, they don’t have millions of DAUs in a coding agent to train from, and they probably have less money, less experience, fewer top tier researchers, and fewer resources for experiments. They are an underdog without a doubt.
None of that means they shouldn’t release their model.
Says who? We know Grok does at the least. They admitted it openly.
seems like they are aiming to provide both inference and RLaaS for american companies and western govts. even if they never fully beat deepseek if they get close enough the fact that they're American will help them close deals
I really enjoyed reading the log book from the training of OPT-175B at Meta… I guess it’s all classified info but it’d be fun to read a blog post about the crazy day to day issues you run into when doing things at this scale :)
The reason I ask is even though it can take hundreds or thousands of contractors to teach a model a certain behavior, wonder if they really only need to do it once for each desired behavior? (of course, future models might expand and refine this previous training) Because if that is the case, then wow, then future models can really expand their capabilities really very fast.
...Am wondering if they somehow record a training session so they can play it back whenever they need to train a new model with the same info? Or maybe the new models can just use distillation from the old model to relearn the old behaviors?
Regarding your second question, I think if they want to further post-train a model using new data, they wouldn't feel the need to re-train it again on the data that it has already being trained on. But you never know. If the model is a completely new pre-trained base model, then they could either train the model using all the data and/or use a previous model to teach it. There is definitely a bunch of tricks they do to evaluate the models and check the performance or whatever their recipe is. It's really up to what the engineers would prefer. But you get the core idea, the models are not suddenly coming up with how to use the Blender on their own, they are explicitly being trained on a dataset curated by a professional that teaches the model how to use Blender. Surely there is another aspect that if the model gets better at coding, then it also helps it become better at Blender, and you have that transfer learning. However, there is no emergence or a deity popping up. But you get people who were evaluating theses models on blender use and suddenly seeing the model ace their tasks and they think they are dealing with a super-intelligence. They then undergo an AI psychosis once they try and extrapolate the (super)-exponential improvement in that one task over the next few months and across all other domains.
Regarding your third question, I already answered at it. But, when it comes to training, they definitely freeze the weights after each run just in case an issue arises and they need to address it (a GPU not working or the loss value blowing up).
It's true that no benchmark communicates the whole picture, and we won't really know how it behaves until weights are out, but the performance here doesn't seem particularly groundbreaking just based on the benchmark.
1: https://en.wikipedia.org/wiki/Qwen
I am more interested with a SOTA Frontier 8B-10B model. Is this even possible?
It's great to see a company that acknowledges it still needs improvement instead of making false claims.
participation awards are not helpful.
Access is currently limited. We'll contact you if early access becomes available.
It's interesting how the industry converged to this very term, given that very less work is being done by horses since quite a while.
https://github.com/inclusionAI/Ling
BTW, this is also an AI lab from China
Where do I get the data?
I mean, this many models. They have to start somewhere.
e.g. fineweb dataset is 50TB https://huggingface.co/datasets/HuggingFaceFW/fineweb
I recommend checking papers from Datalogy, Nvidia Nemotron, Ai2 (Ollmo, Tulu, ...) and the recent model from Aleph Alpha if you want to learn more.