> it runs up to 1.7× faster than comparable MoEs on-device, with 184 tokens/s on an Apple M5 Max and up to 21K output tokens/s on a single NVIDIA B300 GPU
Wow, those are seriously impressive numbers for how intelligent this model seems to be!
One thing that this article doesn't clarify, and that I guess I really should look into, is what exactly AutomationBench tests for as compared to other benchmarks. The standout performance on AutomationBench is interesting to see, but I don't know what it really means because the model seems to score relatively close to other local models on every other benchmark.
I'd guess it's mostly just an artifact of the strength of their Agentic RL environments for those specific domains. If they had the same kind of access to massive amounts of institutional software knowledge, compute infrastructure, and hundreds of millions of users dropping them endless highly technical agent traces, there may have been similar disparities on the agentic benchmarks in other fields.
It's possible that they may have generally capped out for some reason... but, I'd rather be bullish on a new American AI lab making it to first place among western open-weight models that can run on a common consumer device. (Closest in size that's better across that range of benchmarks is Inkling-Small I think, at 276B params, unless I'm forgetting someone.)
Inkling last I used it was pretty terrible. I've never paid for a model that got stuck in a token puke loop and couldn't do tool calls until that day. Even the dated qwen models from almost a year ago 1/10th the size are a better user experience than the inkling model I piloted. Hoping they improve, sincerely, but not a good experience at all.
What does it matter what country an open weight model comes from? Please don't say you tried asking about tienamen square to a model 2 years ago and this was so scary you couldn't use it like 85% of people with this take have said...
Very cool. I’ve been on Gemma 4 12B because of my limited RAM. There’s already an Unsloth quantization of this model so I’m downloading that and will check its performance in the morning. Someone else in this thread recommended Ternary-Bonsai-27B, which I’ll also try.
Just also wanted to note that this model has a questionably-enforceable non-commercial use clause in its license. I really wish they had consulted with an IP lawyer, it’s very confusing.
Don't waste time on bonsai unless you have like 16gb or less ram (ram, not vram). Although I'm pretty sure you could still run this model well on 16+8, and qwen3.6 35b-a3b runs well on ~32gb combined ram at q4. Just offload most of the experts to cpu/ram if you have limited vram budget. With the moe models it doesn't hurt performance too much.
But bonsai is basically trash on real agentic coding work. I've tried both bonsai models, all the quants, messed with the hyperparameters, tried to fine tune (didn't succeed), and it just kept looping or failing to finish tasks. Qwen3.6 35b-a3b is my current go to for agentic work for any systems under 24gb vram, although honestly I'm not sure the dense qwen3.8 27b is meaningfully better. The main difference I see is it does a lot more testing attempts and burns a ton of tokens + compactions, so every task it takes 1.2-2x as long as 35b-a3b does. But the quality of the results is comparable.
Thank you, very interesting. I also often use qwen 35b-a3b, when doing things locally, as I only have ~4.5GB free VRAM. My other goto is gemma 4 26b-a4b, which has faster prompt processing and the same token generation speed.
I've experimented with kolibri1 (78B-a3.5b), but despite the similar size in active parameters it is much slower. Even though the experts fit completely in main memory, as for the other models. I haven't tuned llama.cpp for it as much, so maybe there is still something to gain?
If Qwen had released a Qwen3.8 MoE of similar size, with a naive extrapolation its capabilities would roughly be similar to this model. So basically someone has caught up to Qwen (albeit Qwen from 1-2 months ago) in parameter efficiency.
I don't agree with that conclusion. I'm looking at the benchmarks page for LFM2.5-24B-A2B, and it appears to get a worse score on SWE-Bench-Verified (by a lot) when compared to Qwen 3.6 35B-A3B. It only clearly beats Qwen 3.6 35B on AutomationBench, ToolSandbox, and LiveClaw.
At least for my use cases, Qwen 3.6 35B-A3B appears better than this new release. An MoE for Qwen 3.8 would be even better than that!
Would you call Qwen 3.6 35B "similar size" with ~50% more parameters (both active and total) though? I believe what GP meant that if there was a Qwen 3.8 24B-A2B, it would roughly be as intelligent as this one.
yeah it's kind of the point. they're pedagogical tools. they give you the full recipe for building it from scratch or fine-tuning it beyond what it is.
but i'm prepared to be surprised, maybe they'll conjure up something standalone useful this time
I think their entire pitch about local AI is best demonstrated with their own model that they can optimise for it, and the shape of this model seems optimised for the smaller RTX chips. Indeed, this model appears to be part of their launch picture:
Maybe someone knows what order of cost a release like this requires in real terms of hardware and energy spend?
The cost over the diff from the moment of beginning training to the end. It's iterative from a prior model so it starts there and becomes this release after spending occurs I presume.
For example, is liquid spending millions in energy in this case and the cost is amortized over successive training runs over the initial datacenter spend? What if it's just eating cloud costs and has no dedicated hardware, that sort of thing. I don't particularly mean this model and liquid specifically. Just for this class of smaller fast on-device range of model what is the order of the training cost in real terms?
Maybe its short sighted to ask it as if it's a singular training run and is perhaps the end state of many variations of testing thus all of the R&D training runs would have to be included to to inform what became the final release. Seems like this should be taken into consideration. If anyone happens to know it'd be appreciated!
Training a mid-size model like this can take anywhere between $500k to like $5m. Liquid is spending money on GPU-hours (not sure if you'd call that "energy"). The amortization question is harder to answer without knowing their financials.
The license doesn't really make much sense to me, because if you're already at $10m revenue, you can probably train your own (better, more tailor-suited) model, or at the very least raise money to do so.
I asked it to implement the simplest todo in ghostty codebase and it failed badly - kept going in circles fixing one compile error and introducing another. Qwen 3.8 was slower (40 tok/s compared to 75) but did get it done.
in my experience it's stronger for terminal use, like chaining bash calls, monitoring background tasks, etc. rather than for straight up coding/swe work. Good with web search and file search rather than just editing software.
In a private benchmark I have, it performs much, much worse than Qwen 35B A3B for general knowledge. Still, probably beats Qwen 9B at much higher speed.
200-250 is the rate I've seen my context window fill up during longer tasks with Claude. The 40-80 figure is what Google's AI mode says, but I don't know where it got that from.
I've been messing around low computing power hardware to run models locally and to me 100 is already plenty fast. You most likely won't notice a huge difference between 184 and 250, unless you're running some serious multi-age agent circus.
Another data point: I got 94.74 tokens/s on an M4 Pro Mac mini 64GB with the simonw prompt "Generate an SVG of a pelican riding a bicycle". TTFT was 0.37s. The model thought for 40.75s before answering.
>How many tokens/s do you need for a coding agent to feel like Claude or codex?
it's kind of a loaded question, this model gets a lot of tok/s but reasoning is forced on; the time-to-answer is really a combination of things like "how much does this thing want to over-reason" and "how much does this thing like tool use?" .
You are right, they are not exactly the same thing, just wanted to have a reference point
When running locally usually I have to trade off speed for a better model. I can run decently big models at about 2tk/s, or a pretty dumb one at 40-100tk/s
AFAIU prefill speed makes a bigger difference in agentic coding workflows as the harness likely spends more time reading than creating new code. Qwen in particular spends a lot of time reasoning so it does make a larger impact there
> it runs up to 1.7× faster than comparable MoEs on-device, with 184 tokens/s on an Apple M5 Max and up to 21K output tokens/s on a single NVIDIA B300 GPU
Wow, those are seriously impressive numbers for how intelligent this model seems to be!
One thing that this article doesn't clarify, and that I guess I really should look into, is what exactly AutomationBench tests for as compared to other benchmarks. The standout performance on AutomationBench is interesting to see, but I don't know what it really means because the model seems to score relatively close to other local models on every other benchmark.
It scores only a 6 on the "Artificial Analysis Intelligence Index". That makes it basically useless.
I'd guess it's mostly just an artifact of the strength of their Agentic RL environments for those specific domains. If they had the same kind of access to massive amounts of institutional software knowledge, compute infrastructure, and hundreds of millions of users dropping them endless highly technical agent traces, there may have been similar disparities on the agentic benchmarks in other fields.
It's possible that they may have generally capped out for some reason... but, I'd rather be bullish on a new American AI lab making it to first place among western open-weight models that can run on a common consumer device. (Closest in size that's better across that range of benchmarks is Inkling-Small I think, at 276B params, unless I'm forgetting someone.)
Inkling last I used it was pretty terrible. I've never paid for a model that got stuck in a token puke loop and couldn't do tool calls until that day. Even the dated qwen models from almost a year ago 1/10th the size are a better user experience than the inkling model I piloted. Hoping they improve, sincerely, but not a good experience at all.
What does it matter what country an open weight model comes from? Please don't say you tried asking about tienamen square to a model 2 years ago and this was so scary you couldn't use it like 85% of people with this take have said...
It is very fast on a 4090, but also very stupid.
Which makes for a dangerous combination, because it will delete prod, and it will do so at 250t/s.
If your AI agent has the privileges to delete prod, the problem is not that it is stupid.
Very cool. I’ve been on Gemma 4 12B because of my limited RAM. There’s already an Unsloth quantization of this model so I’m downloading that and will check its performance in the morning. Someone else in this thread recommended Ternary-Bonsai-27B, which I’ll also try.
Just also wanted to note that this model has a questionably-enforceable non-commercial use clause in its license. I really wish they had consulted with an IP lawyer, it’s very confusing.
Don't waste time on bonsai unless you have like 16gb or less ram (ram, not vram). Although I'm pretty sure you could still run this model well on 16+8, and qwen3.6 35b-a3b runs well on ~32gb combined ram at q4. Just offload most of the experts to cpu/ram if you have limited vram budget. With the moe models it doesn't hurt performance too much.
But bonsai is basically trash on real agentic coding work. I've tried both bonsai models, all the quants, messed with the hyperparameters, tried to fine tune (didn't succeed), and it just kept looping or failing to finish tasks. Qwen3.6 35b-a3b is my current go to for agentic work for any systems under 24gb vram, although honestly I'm not sure the dense qwen3.8 27b is meaningfully better. The main difference I see is it does a lot more testing attempts and burns a ton of tokens + compactions, so every task it takes 1.2-2x as long as 35b-a3b does. But the quality of the results is comparable.
Thank you, very interesting. I also often use qwen 35b-a3b, when doing things locally, as I only have ~4.5GB free VRAM. My other goto is gemma 4 26b-a4b, which has faster prompt processing and the same token generation speed.
I've experimented with kolibri1 (78B-a3.5b), but despite the similar size in active parameters it is much slower. Even though the experts fit completely in main memory, as for the other models. I haven't tuned llama.cpp for it as much, so maybe there is still something to gain?
This seems like a big deal, no?
If Qwen had released a Qwen3.8 MoE of similar size, with a naive extrapolation its capabilities would roughly be similar to this model. So basically someone has caught up to Qwen (albeit Qwen from 1-2 months ago) in parameter efficiency.
I don't agree with that conclusion. I'm looking at the benchmarks page for LFM2.5-24B-A2B, and it appears to get a worse score on SWE-Bench-Verified (by a lot) when compared to Qwen 3.6 35B-A3B. It only clearly beats Qwen 3.6 35B on AutomationBench, ToolSandbox, and LiveClaw.
At least for my use cases, Qwen 3.6 35B-A3B appears better than this new release. An MoE for Qwen 3.8 would be even better than that!
Would you call Qwen 3.6 35B "similar size" with ~50% more parameters (both active and total) though? I believe what GP meant that if there was a Qwen 3.8 24B-A2B, it would roughly be as intelligent as this one.
Smallest Qwen Q3 quant is the same size as LFM 25B Q4 ≈ 14 GB, but much better at coding and following my instructions.
Way more excited for Nvidia's A9B model coming out tomorrow.
Why? Their models are useful as a starting place for doing your own fine tuning, but they have never stood well on their own.
they always undertrain the model
yeah it's kind of the point. they're pedagogical tools. they give you the full recipe for building it from scratch or fine-tuning it beyond what it is.
but i'm prepared to be surprised, maybe they'll conjure up something standalone useful this time
They have an incentive to produce a more finished model to go with the launch of the RTX Spark laptops, I guess.
Not really - you pay for the hardware. Everyone else already is building models for Spark hardware.
I think their entire pitch about local AI is best demonstrated with their own model that they can optimise for it, and the shape of this model seems optimised for the smaller RTX chips. Indeed, this model appears to be part of their launch picture:
https://blogs.nvidia.com/blog/local-ai-rtx-spark-laptops-sof...
Maybe someone knows what order of cost a release like this requires in real terms of hardware and energy spend?
The cost over the diff from the moment of beginning training to the end. It's iterative from a prior model so it starts there and becomes this release after spending occurs I presume.
For example, is liquid spending millions in energy in this case and the cost is amortized over successive training runs over the initial datacenter spend? What if it's just eating cloud costs and has no dedicated hardware, that sort of thing. I don't particularly mean this model and liquid specifically. Just for this class of smaller fast on-device range of model what is the order of the training cost in real terms?
Maybe its short sighted to ask it as if it's a singular training run and is perhaps the end state of many variations of testing thus all of the R&D training runs would have to be included to to inform what became the final release. Seems like this should be taken into consideration. If anyone happens to know it'd be appreciated!
Training a mid-size model like this can take anywhere between $500k to like $5m. Liquid is spending money on GPU-hours (not sure if you'd call that "energy"). The amortization question is harder to answer without knowing their financials.
The license doesn't really make much sense to me, because if you're already at $10m revenue, you can probably train your own (better, more tailor-suited) model, or at the very least raise money to do so.
I asked it to implement the simplest todo in ghostty codebase and it failed badly - kept going in circles fixing one compile error and introducing another. Qwen 3.8 was slower (40 tok/s compared to 75) but did get it done.
> Qwen 3.8 was slower
Is that Qwen 3.8 27B? Because yes, you'd expect a plain 27B model to be better than a 24B-A2B MoE model all else being equal.
(All else isn't equal though of course - Qwen models are very, very good at coding)
Gemma-4 26BA4B would be a more similar comparison.
That's not surprising. Even Liquid's own documentation states that coding isn't this model's strong suit. Its strength lies in tool calling.
Have you tried Gemma-4 26BA4B? I feel like it is underrated for tasks like that, but maybe I've just been lucky with my use cases.
in my experience it's stronger for terminal use, like chaining bash calls, monitoring background tasks, etc. rather than for straight up coding/swe work. Good with web search and file search rather than just editing software.
In a private benchmark I have, it performs much, much worse than Qwen 35B A3B for general knowledge. Still, probably beats Qwen 9B at much higher speed.
Same. Asked for some questions and it put out some old knowledge as well.
fast is fast (18tk/s on a Intel Core 7 155H laptop), but quality is something else
Is its purpose general knowledge?
How about if you let use with web search, tool calls, and RAG? Does iimprove?
I'm running a Ryzen 7 8745HS with Radeon 780M (27GB RAM), using Ollama's Vulkan backend.
I get ~45 tokens/s generation and ~680 tokens/s prompt processing.
The problem is ollama is not caching the prompt, I get -> forcing full prompt re-processing due to lack of cache data
So this wouldn't work in its current state in an agentic loop.
Friends don't let friends use Ollama: https://sleepingrobots.com/dreams/stop-using-ollama/
What a pain to be forced to use dark mode...
Ollama is generally just not great software. Use something else and you likely won't have that issue.
i'm gettin between 200-320 tok/s on a 5070ti with 131k context on llamacpp and no tuning.
very nice.
I can't speak to the intelligence of it yet.
these small agentic-focused models are a lot of fun to play with if you haven't yet.
Love seeing the effort Liquid is putting into local models. I’m looking forward to a lot of progress here in the next 5 years
> with 184 tokens/s on an Apple M5 Max
That seems pretty decent. How many tokens/s do you need for a coding agent to feel like Claude or codex?
200-250 is the rate I've seen my context window fill up during longer tasks with Claude. The 40-80 figure is what Google's AI mode says, but I don't know where it got that from.
I've been messing around low computing power hardware to run models locally and to me 100 is already plenty fast. You most likely won't notice a huge difference between 184 and 250, unless you're running some serious multi-age agent circus.
Another data point: I got 94.74 tokens/s on an M4 Pro Mac mini 64GB with the simonw prompt "Generate an SVG of a pelican riding a bicycle". TTFT was 0.37s. The model thought for 40.75s before answering.
(The resulting pelican was underwhelming.)
Claude code is 60-80 tps, codex varies by model but mostly 40-50 tps.
>How many tokens/s do you need for a coding agent to feel like Claude or codex?
it's kind of a loaded question, this model gets a lot of tok/s but reasoning is forced on; the time-to-answer is really a combination of things like "how much does this thing want to over-reason" and "how much does this thing like tool use?" .
You are right, they are not exactly the same thing, just wanted to have a reference point
When running locally usually I have to trade off speed for a better model. I can run decently big models at about 2tk/s, or a pretty dumb one at 40-100tk/s
fwiw this thing puts out results very fast even on max reasoning.
the results i've gotten so far have been a little braindead though. still playing around. bad luck with traversing filesystems remotely so far.
it seems to want to make concluding decisions very fast regardless of reasoning levels.
AFAIU prefill speed makes a bigger difference in agentic coding workflows as the harness likely spends more time reading than creating new code. Qwen in particular spends a lot of time reasoning so it does make a larger impact there
I’d say only 65
Does anyone know the order of magnitude licensing cost for the model for businesses over 10m in revenue?
The license is generrous (free under <$10m) Pricing confusing when it shows API pricing 1st
Always good to hear for on-device suited model, but always cannot sure how it real-performance in real task..
It is a good model for on-device work.