88 comments

[ 2.9 ms ] story [ 65.8 ms ] thread
The big story here is the encoder-free part, which I still don't fully understand.

> Vision: We replaced Gemma 4’s vision encoder with a lightweight embedding module consisting of a single matrix multiplication, positional embedding and normalizations.

That's technically encoding, just without using a dedicated model for it like SigLIP? The Developer's Guide elaborates, it's still a 35M layer which I am curious is robust enough. https://developers.googleblog.com/gemma-4-12b-the-developer-...

> Small enough to run locally on consumer laptops with 16GB of RAM, it unlocks powerful multimodal and agentic experiences right on your machine.

I am assuming that involves quantization, which due to the quality loss makes that statement somewhat misleading IMO.

Wow Google is becoming the new pre Llama 4 Meta when it comes to releasing open weights models.
What's Google's business case for releasing open models? Don't get me wrong, I am grateful and appreciative of these releases. I'm trying to understand how it fits into their bigger picture as a for profit company? Are they not helping competitors build on the novel technology they have developed?

Is it simply goodwill and/or marketing? Or am I missing something strategic?

(comment deleted)
How does it compare with e4b, aside from being larger?
This is a pretty good update. The demo video is a bit funny though - the tester asks to turn the release into bullet points. okay, the model obliges. then the tester says draft an email with this content. BAM! the LLM turns the content from bullets to passages even though it was not asked and it undid the last good thing that it did. i am not sure if it's an email etiquette to not put bullets in the email.
Is this Mac only? Or is that an Ollama issue that it only supports this release of models on Mac? It seems like every tag with the MLX badge is only supported on Mac[0], and that includes all of the tags in this release.

[0] https://ollama.com/library/gemma4/tags

Edit: MLX being Mac-only is independent of the model being MLX (and therefore Mac) only. The latter is what I am asking about.

> Novel unified architecture: No multimodal encoders. The vision and audio inputs flow directly into the LLM backbone.

I would be interested in how this actually works. I couldn't find a description of the model architecture (and I did check the links in the Google blog)

What are the use cases for these small models? Is there anyone using models of this scale in their daily life who could share their experience?
Yes, all my emails gyer sorted out by a finetuned gemma. There are turned into images passes to the model, as multimodal is so practical.
I moreso run other small special purpose models like Whisper, SAM, Matcha, CLIP etc. and then do contextual correction passes with models like this.

Think almost like unix pipelines, have used it for many workflows.

I can’t help but wonder if this is the basis of the model they’ve helped tune for Apple.
I do enjoy the immediate out of touch signaling with the "runs on your 16gb vram laptop" line. Because everyone has a laptop with 16gb vram, or can just pop out and buy a new one, right?
They already provide E2B and E4B that run on (much) smaller devices, including tablets and phones. This fills the gap in the middle. The bigger Gemma 4 models are excellent for their size, but at 8-bit quantization they need about 64GB of VRAM or unified memory. 48GB for 6-bit. Any lower quantization than that, they start to get notably dumber. So, a 12B is interesting for that middle ground.
Surely they must know the current hurdles, but clearly they know that all the relevant people are monitoring the market for the proper hardware to get and 16GB will be an entry point.
Quite a niche release. The MoE outperforms it on score and will likely be faster thanks to lower active weights. So this really only makes sense for specific ram constrained applications that can’t fit a quantized MoE
Quite aside from the architectural changes, I suppose this is the answer to why Google had such a glaring hole in the (pretrained) Gemma4 model lineup between the Gemma4 4b and Gemma4 26b models!

A model that comfortably fits in 16GB of VRAM (allowing room for context) is a welcome upgrade.

I don’t see the download in lm studio
using an embedder instead of a decoder is quite clever. Not sure who came up with that first but it's a cool idea.
Am I missing something or are the Ollama versions of this (https://ollama.com/library/gemma4/tags) text-only for now?
Highly recommend just dropping Ollama. You can download binary releases of llama.cpp for every platform and run them trivially in 5 seconds. Ollama serves no purpose other than to take open source work and rebadge as its own, while providing inferior functionality
(comment deleted)
Is there some place where we can try it before downloading the gigabytes of weights?
Is there a paper on this?

I'm curious how they pre-trained it... I feel like it must have had audio/image output that they chopped off.

I wonder how hard it would be to add it back on.

I'm waiting for FP8 quant, preferably from Google.
A small dense multimodal model with audio support, interesting.

Wait, *Excluding Chinese language.

This is ... curious.

P.S. Where is gemma 4 124b?

Are there qwen or minimax or other open weight models of same hardware requirements that outperform this?
I ran the Q4 quant (used with llama.cpp) though my "minesweeper" vibe-coding benchmark: https://senko.net/vibecode-bench/2026/minesweeper-gamma-4-12...

The result is decent, but it had a few bizzare/trivial syntax errors I had to fix manually: it would do an extra closing bracket or paren a few times, and wanted to separate function definitions with comma. Not sure what that was about, but otherwise the output run just fine.

So, with those qualifiers, I think it's a decent local coding model. It roughly compares with GPT-4.1 (!!), released 14 months ago, on the output: https://senko.net/vibecode-bench/2025/minesweeper-gpt-4.1.ht... (actually I'd call it better, but those syntax errors...)

I ran the quantized version (4-bit GGUF) on my consumer-grade card with 12G of VRAM and got 5t/s for output. Not for interactive use for coding, but fairly capable model.

To me, it's fascinating how much progress we got in over a year. GPT-4.1 was considered an extremely capable coding model. Now we got something with 12B of params performing roughly the same (in this specific benchmark, disclaimers, etc).

Lists of various models I tested: https://senko.net/vibecode-bench/

Models this small and this capable bode really well for the usefulness of a PC like the RTX Spark that Nvidia/Microsoft announced this week. 128GB of unified memory will likely be more than sufficient for effective local agentic coding, even if SOTA cloud models will still be even better.

Up until this point, I've found the cost/value to unequivocally favor using a cloud subscription, but I would be lying if I didn't worry that one day OpenAI is going to increase the price for my subscription by 5-10x. I rely on these tools enough that if there is a real viable local option, I'm going to take it.

>The result is decent, but it had a few bizzare/trivial syntax errors I had to fix manually

Can you instruct it to use a lsp?

We are really getting close to singularity - the pace of LLM improvement is constantly accelerating.
I've heard the assertion that the Gemma 4 models don't do well with lower quantization. I wonder if the "bizzare/trivial" syntax errors would go away at Q8?
> it would do an extra closing bracket or paren a few times

I had this with Gemini: in the middle of a C++ program it once said RParen instead of using )

It was easy to fix of course, but it makes you question what is going on inside its head.

The Unsloth 8bit quant seems to manage this task without any syntax errors.
Last time I tried Gemma 4 (26B-A4B) its memory usage would balloon and consume all of my swap until my machine died.

Qwen 3.6 on the other hand barely uses any memory at all for its KV cache.