Gemma TL @GoogleDeepMind
Paris, France
Joined September 2014
- Tweets221
- Following481
- Followers2.1K
- Likes513
Pinned Tweet
Gemma 4 is out! 🔥 Once again, our team has delivered best-in-class open weights models, fully available for research and commercial use!
Robert Dadashi retweeted
Very much agree with this letter.
3 years ago we started Gemma aligned with this vision, that building a thriving open weight model ecosystem was critical and likely a huge component of our future industry.
Today we continue to invest and push Gemma to be maximally useful for developers, and I’m personally very bullish on the open ecosystem it enables.
Thanks @JensenHuang and also great to see you here!
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
Robert Dadashi retweeted
Our goal in the Gemma team is to ship models that are useful by generalizing to unseen tasks. Hence, we are extremely strict about not doing anything that would target specific benchmarks instead of teaching the model broad capabilities. It's great to see this reflected here.
Qwen3.5 27B vs Gemma 4 31B IT
Here is why I think Gemma 4 31B wins:
Benchmark scores of both models are very close.
Generalization is not.
I reused NVIDIA’s CoDeC metric to probe "benchmark affinity".
Officially, CoDeC > 80 suggests contamination.
I've never seen this. But even below that, it can still reveal how benchmark-affine a model is.
Higher CoDeC scores mean the model is more "comfortable" with the benchmark.
So the ideal profile is:
high benchmark scores + low CoDeC.
By that lens, Gemma 4 looks clearly stronger than Qwen3.5.
Gemma 4 likely generalizes better to unseen tasks.
I observed the same pattern earlier with Gemma 3 vs Qwen3.
I wrote about this a few months ago:
kaitchup.substack.com/p/did-…
Robert Dadashi retweeted
Proud to announce the release of Gemma 4 💎 !! Better than Gemma 3 across the board on benchmarks+small models with audio
We bring you significant vision improvememts including variable aspect ratio, variable resolution image processing & better spatial understanding!
Robert Dadashi retweeted
Gemma 4 is finally here!💎✨
Incredibly proud to be part of the team behind these models and I can't wait to see what you build with them.
Robert Dadashi retweeted
As part of the Gemma 4 release, we're launching Agent Skills: an Android app experience where you can import different skills and have Gemma 4 E2B reason and use the skills!
Running entirely in the phone, available in the Google PlayStore. Try it now!
Robert Dadashi retweeted
Gemma 4 looks great on paper, just what many people need, but mostly I'm focused on the questions around what makes an open model succeed in the long term.
We desperately need to build a field of research on "what makes a finetunable model" if we want the open model economy to succeed. Too much of that research happens behind closed doors of companies.
Plus, 240 days after The ATOM Project was released, I share reflections on the recovery of American models, and what that means for the next phase.
interconnects.ai/p/gemma-4-a…
Robert Dadashi retweeted
Good account to follow: @googlegemma
Robert Dadashi retweeted
💎💎💎💎 Huge news today: we're launching #Gemma4! Our most capable open models yet.
🔓 Apache 2.0: Complete flexibility and digital sovereignty
🧠 Advanced Reasoning: Multi-step planning and deep logic
🛠️ Agentic Workflows: Native support for function-calling and structured JSON
👁️👂 Native Multimodal: Vision and audio input natively supported on edge models
🧑💻 Code generation: unleash local AI code generation
📱 From edge models 2B & 4B...
💻 to Workstation models 26B MoE (3.8B active) & 31B Dense
🏁 Natively trained on over 140 languages to enable inclusive, high-performance applications for a global audience
Read the blog: blog.google/innovation-and-a…
Robert Dadashi retweeted
Gemma 💎💎💎💎
Fantastic models, fantastic team!
Robert Dadashi retweeted
Gemma 4 is out! They are the best open models in the world for their size 💎
Congrats everyone for the hard work, excited to see what people will build with them.
More info at: deepmind.google/models/gemma…
Robert Dadashi retweeted
Really proud of everyone in the team that worked hard for this release: Gemma 4 is here! blog.google/innovation-and-a…
Robert Dadashi retweeted
Gemma 4 is out! 💎💎💎💎
Among many things, I am quite excited that Gemma 4 E2B thinking is roughly matching Gemma 3 27B across benchmarks (screenshot below)!
More info here: deepmind.google/models/gemma…
Gemma is hiring! 🚀🚀🚀
Please apply if you want to work on pre-training and/or designing models at the intersection of multiple modalities (text, audio, vision, video, actions,)
Location: Paris
Apply here: job-boards.greenhouse.io/dee…
Robert Dadashi retweeted
An exciting milestone for AI in science: Our C2S-Scale 27B foundation model, built with @Yale and based on Gemma, generated a novel hypothesis about cancer cellular behavior, which scientists experimentally validated in living cells.
With more preclinical and clinical tests, this discovery may reveal a promising new pathway for developing therapies to fight cancer.
Robert Dadashi retweeted
MLX-VLM v0.2.0 is here 🔥🙌🏽
Or should I rename it to MLX-O since Omni is taken?
We now support Omni models starting with the latest Gemma3n from @GoogleDeepMind and more coming soon.
Thanks to @fleetwood___ , @pcuenq, @osanseviero, @reach_vb and team for helping make this possible!
Get started today:
> pip install -U mlx-vlm
🚀 Gemma 3n is HERE with day-0 MLX support!
Thrilled to have collaborated with @GoogleDeepMind and @huggingface to bring the MLX community instant access to this groundbreaking model.
Why Gemma 3n changes everything:
📹 True Multimodal: text, audio, image & video
⚡ Runs on just a few GB of RAM
🏆 First <10B model to hit 1300+ on @lmarena_ai
Massive thanks to @osanseviero, @reach_vb, @fleetwood___, @pcuenq and Ryan for making this possible! 🙏
Get started in seconds:
> pip install -U mlx-vlm
Try it now and share what you build! 🔥
youtu.be/8-8R2UvUBrc?si=5WzE…
Gemma 3n E4B has the same number of total parameters (8B) as the original Gemma 7B (8B lol). The progress of Gemma over the past 16 months is insane
I’m so excited to announce Gemma 3n is here! 🎉
🔊Multimodal (text/audio/image/video) understanding
🤯Runs with as little as 2GB of RAM
🏆First model under 10B with @lmarena_ai score of 1300+
Available now on @huggingface, @kaggle, llama.cpp, ai.dev, and more
Robert Dadashi retweeted
Very excited to bring amazing capabilities to on device/mobile footprints. Enjoy building with Gemma!
Our open source Gemma models are the most powerful single GPU/TPU models out there! Our latest model Gemma 3n has amazing performance, multimodal understanding, & can run with as little as 2GB of memory - perfect for edge devices - enjoy building at ai.studio !
Our latest Gemma model is finally out 🔥🔥🔥
Our open source Gemma models are the most powerful single GPU/TPU models out there! Our latest model Gemma 3n has amazing performance, multimodal understanding, & can run with as little as 2GB of memory - perfect for edge devices - enjoy building at ai.studio !