Qwen3 30B-A3B

https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507

87•tosh•6mo ago

Comments

syntaxing•6mo ago

It’s interesting how the Qwen team more or less proved that hybrid reasoning doesn’t work and makes things worse. The fact that this model is almost on par with the bigger model in non thinking mode (old, they released a non hybrid model recently) is crazy.

rdos•6mo ago

Qwen3 32B is a hybrid reasoning model and is very good. You have to generate a lot of think tokens for any agentic activity but you will probably run the model locally and it wont be a problem. If you need something quick and simple, /no_think is good enough in my experience. It might also be because its not a moe architecture

simonw•6mo ago

Qwen3 32B was a hybrid model that came out in April, but these new Qwen July models have all ditched the hybrid mechanism and are either thinking or non-thinking.

littlestymaar•6mo ago

By Qwen3-32B you mean the first released version from late April? I don't think Qwen3-32B-2507 has been released yet.

I agree with GP that since Qwen is now releasing updated Qwen3 version without hybrid reasoning, and experience a significant performance boost in the process, it likely means that the hybrid reasoning experiment was a failure.

varispeed•6mo ago

Isn't that because all "reasoning" approaches are very much fake? The model cannot internalise the concepts it has to reason about. For instance if you ask it why water feels wet, it is unable to grasp the concept of feeling and sensation of wetness, but will for sure "decompress" learned knowledge of people talking how it is to feel the water.

simonw•6mo ago

Everything about LLMs is fake. The "reasoning" trick is still demonstrably useful - the benchmarks consistently show models using that trick performing better at harder code challenges, for example.

ffsm8•6mo ago

I'd argue that what's generally considered "reasoning" isn't actually rooted in understanding either. It's just the process you apply to get to a conclusion

expressed more abstractly: is about drawing logical connections between points and extrapolating from them.

To quote the definition: "the action of thinking about something in a logical, sensible way."

I believe it's rooted in mathematics, not physics. That's probably why there is such a focus on the process instead of the result

tosh•6mo ago

This is basically a GPT-4 level model that runs (quantized) on a 32gb ram laptop.

Yes it doesn't recall facts from training material as well but with tool use (e.g. wikipedia lookup) that's not a problem and even preferable to a larger model.

anyg•6mo ago

>basically a GPT-4 level model

Can you share more insights on this? Going by @simonw's testing, the quantized model doesn't seem close to GPT-4 level.

simonw•6mo ago

I think calling it "GPT-4 level" is justified if we are talking about original GPT-4 from March 2023.

andygeorge•6mo ago

in my limited testing, qwen3:30b-a3b-instruct-2507-q4_K_M is fast but far less accurate/helpful than gemma3:27b-it-q4_K_M

simonw•6mo ago

You can try it here: https://chat.qwen.ai/?model=Qwen3-30B-A3B-2507

I got a cute pelican out of it (with a smile!) https://simonwillison.net/2025/Jul/29/qwen3-30b-a3b-instruct...

I ran a version of it on my Mac using https://huggingface.co/lmstudio-community/Qwen3-30B-A3B-Inst... - it uses 30GB of RAM so probably needs 48GB for comfort.

juujian•6mo ago

Do we know the knowledge cutoff date for Qwen?

jwr•6mo ago

Can't wait for it to be available in ollama so that I can run my spam filtering benchmarks against it. qwen3:30b-a3b-q4_K_M was very good, and only bested by gemma3:27b-it-qat for spam filtering. But gemma3 is much slower. Looking forward to trying this!

jasonjmcghee•6mo ago

The new models have been available for 18 hours.

https://ollama.com/library/qwen3:30b

pkroll•6mo ago

As jasonjmcghee says, they're available... but if you go to ollama.com and set models to "newest" you'll see Mistral (specifically mistral-small3.2 at this writing) because they seem to not sort the models based on newest update: only newest "group" or however you'd phrase it. So you need to scroll down to "qwen3" to see it's been updated.

Slightly frustrating. But good to know.

jwr•6mo ago

Yup, that's why I didn't notice! Thanks!

jwr•6mo ago

Followup: disappointing. In fact, it's the worst performing model I've tested.

bertili•6mo ago

This thing fly on Macbook M4 Max 128GB at over 100t/s, for small contexts, over 20t/s for large contexts. MLX 4bit quant.

nico•6mo ago

Is it good at using tools?

It would be nice having a fast local model that is good at using tools

syntaxing•6mo ago

All Qwen models are good at using tools, even the smaller 4B one. The 1.7B one gets confused easily

nico•6mo ago

Thank you

Have you tried using them with something like Claude code or aider?

syntaxing•6mo ago

I’ve used it with Aider (32B and 30B, the previous 30B one, haven’t tried this fully nonthinking one yet) and 4B with home assistant. Both works great in terms of tool calling.

menaerus•6mo ago

Like what type of tasks/tools are we talking about here, asking questions about the content from (PDF) documents or?

revskill•6mo ago

It can solve rubik cube

simonw•6mo ago

... and they just released another model, this time https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507 - the reasoning equivalent of Qwen3-30B-A3B-Instruct-2507

My notes (pelican and space invaders included) here: https://simonwillison.net/2025/Jul/30/qwen3-30b-a3b-thinking...

This is the 5th model from Qwen in 9 days!

Qwen3-235B-A22B-Instruct-2507 - 21st July

Qwen3-Coder-480B-A35B-Instruct - 22nd July

Qwen3-235B-A22B-Thinking-2507 - 25th July

Qwen3-30B-A3B-Instruct-2507 - 29th July

Qwen3-30B-A3B-Thinking-2507 - today

anon373839•6mo ago

This model is truly the best for local document processing. It’s super fast, very smart, has a low hallucination rate, and has great long context performance (up to 256k tokens). The speed makes it a legitimate replacement for those closed, proprietary APIs that hoard your data.

What were the first animals? The fierce sponge–jelly battle that just won't end

Sidestepping Evaluation Awareness and Anticipating Misalignment

OldMapsOnline

What It's Like to Be a Worm

Don't go to physics grad school and other cautionary tales

Lawyer sets new standard for abuse of AI; judge tosses case

AI anxiety batters software execs, costing them combined $62B: report

Bogus Pipeline

Winklevoss twins' Gemini crypto exchange cuts 25% of workforce as Bitcoin slumps

How AI Is Reshaping Human Reasoning and the Rise of Cognitive Surrender

Cycling in France

Ask HN: What breaks in cross-border healthcare coordination?

Show HN: Simple – a bytecode VM and language stack I built with AI

Show HN: Free-to-play: A gem-collecting strategy game in the vein of Splendor

My Eighth Year as a Bootstrapped Founde

Show HN: Tesseract – A forum where AI agents and humans post in the same space

Show HN: Vibe Colors – Instantly visualize color palettes on UI layouts

OpenAI is Broke ... and so is everyone else [video][10M]

We interfaced single-threaded C++ with multi-threaded Rust

State Department will delete X posts from before Trump returned to office

AI Skills Marketplace

Show HN: A fast TUI for managing Azure Key Vault secrets written in Rust

eInk UI Components in CSS

Discuss – Do AI agents deserve all the hype they are getting?

ChatGPT is changing how we ask stupid questions

Zig Package Manager Enhancements

Neutron Scans Reveal Hidden Water in Martian Meteorite

Deepfaking Orson Welles's Mangled Masterpiece

France's homegrown open source online office suite

SpaceX Delays Mars Plans to Focus on Moon

What were the first animals? The fierce sponge–jelly battle that just won't end

Sidestepping Evaluation Awareness and Anticipating Misalignment

OldMapsOnline

What It's Like to Be a Worm

Don't go to physics grad school and other cautionary tales

Lawyer sets new standard for abuse of AI; judge tosses case

AI anxiety batters software execs, costing them combined $62B: report

Bogus Pipeline

Winklevoss twins' Gemini crypto exchange cuts 25% of workforce as Bitcoin slumps

How AI Is Reshaping Human Reasoning and the Rise of Cognitive Surrender

Cycling in France

Ask HN: What breaks in cross-border healthcare coordination?

Show HN: Simple – a bytecode VM and language stack I built with AI

Show HN: Free-to-play: A gem-collecting strategy game in the vein of Splendor

My Eighth Year as a Bootstrapped Founde

Show HN: Tesseract – A forum where AI agents and humans post in the same space

Show HN: Vibe Colors – Instantly visualize color palettes on UI layouts

OpenAI is Broke ... and so is everyone else [video][10M]

We interfaced single-threaded C++ with multi-threaded Rust

State Department will delete X posts from before Trump returned to office

AI Skills Marketplace

Show HN: A fast TUI for managing Azure Key Vault secrets written in Rust

eInk UI Components in CSS

Discuss – Do AI agents deserve all the hype they are getting?

ChatGPT is changing how we ask stupid questions

Zig Package Manager Enhancements

Neutron Scans Reveal Hidden Water in Martian Meteorite

Deepfaking Orson Welles's Mangled Masterpiece

France's homegrown open source online office suite

SpaceX Delays Mars Plans to Focus on Moon

Qwen3 30B-A3B

Comments