Flagship open-weight release days are always exciting. Was just reading through the Gemma 4 reports, configs, and code, and here are my takeaways: Architecture-wise, besides multi-model support, Gemma 4 (31B) looks pretty much unchanged compared to Gemma 3 (27B). Gemma 4 maintains a relatively unique Pre- and Post-norm setup and remains relatively classic, with a 5:1 hybrid attention mechanism combining a sliding-window (local) layer and a full-attention (global) layer. The attention mechanism itself is also classic Grouped Query Attention (GQA). But let’s not be fooled by the lack of architectural changes. Looking at the benchmarks, Gemma 4 is a huge leap from Gemma 3. This is likely due to the training set and recipe. Interestingly, on the AI Arena Leaderboard, Gemma 4 (31B) ranks similarly to the much larger Qwen3.5-397B-A17B model. But as I discussed in my model evaluation article, arena scores are a bit problematic as they can be gamed and are biased towards human (style) preference. If we look at some other common benchmarks, which I plotted below, we can see that it’s indeed a very clear leap over Gemma 3 and ranks on par with Qwen3.5 27B. Note that there is also a Mixture-of-Experts (MoE) Gemma 4 variant that is slightly smaller (27B with 4 billion parameters active. The benchmarks are only slightly worse compared to Gemma 4 (31B). I omitted the MoE architecture in the figure below because the figure is already very crowded, but you can find it in my LLM Architecture Gallery. Anyways, overall, it's a nice and strong model release and a strong contender for local usage. Also, one aspect that should not be underrated is that (it seems) the model is now released with a standard Apache 2.0 open-source license, which has much friendlier usage terms than the custom Gemma 3 license.
OPEN SOURCE
-
Google Releases Gemma 4: Open Source AI Models
By
–
🔴 ¡GOOGLE LIBERA GEMMA 4!
— Carlos Santana (@DotCSV) 2 avril 2026
Cuatro versiones abiertas muy interesantes:
👉 31B Dense y 26B MoE: rendimiento equivalente a alternativas más grandes en tamaños muy accesibles!
👉 E4B y E2B: ligeros con procesamiento en tiempo real de texto, visión y audio!pic.twitter.com/aGhWF3zXFg¡GOOGLE LIBERA GEMMA 4! Cuatro versiones abiertas muy interesantes: 31B Dense y 26B MoE: rendimiento equivalente a alternativas más grandes en tamaños muy accesibles! E4B y E2B: ligeros con procesamiento en tiempo real de texto, visión y audio!
-
Gemma 4 Models Tested: Pelicans Generated via LM Studio and Gemini API
By
–
Pelicans for Gemma 4 E2B, E4B, 26B-A4B and 31B – the first three generated on my laptop via LM Studio, the 31B was broken on my laptop so I ran it via the Gemini API instead simonwillison.net/2026/Apr/2…
-
Action-to-Action Flow Matching: Ultra-Fast Robot Control Method
By
–
What if real-time robot control didn't have to wait for slow, iterative action generation? MARS Lab at Nanyang Technological University (Jindou Jia et al.) introduces Action-to-Action Flow Matching (A2A). This novel method uses a robot's own historical actions to directly predict the next move, skipping the slow, random noise sampling of traditional diffusion models. A2A enables lightning-fast, single-step action generation (0.56 ms!), vastly outperforming existing methods in speed, training efficiency, robustness to visual noise, and generalization to unseen configurations. It even shows versatility in video generation! Action-to-Action Flow Matching Website: lorenzo-0-0.github.io/A2A_Fl… arXiv: arxiv.org/pdf/2602.07322 Code: github.com/JIAjindou/A2A_Flo… Our report: mp.weixin.qq.com/s/mrSUcVLUA… 📬 #PapersAccepted by Jiqizhixin
→ View original post on X — @jiqizhixin, 2026-04-02 18:30 UTC
-
LLM Trained on Pre-1900 Text Discovers Modern Physics
By
–
I trained an LLM from scratch on pre-1900 text to see if it could come up with quantum mechanics and relativity.
— Michael Hla (@hla_michael) 2 avril 2026
While the model is too small to do meaningful reasoning, it has glimpses of intuition.
When given observations from past landmark experiments, the model can declare… pic.twitter.com/qVjUF6bvtpI trained an LLM from scratch on pre-1900 text to see if it could come up with quantum mechanics and relativity. While the model is too small to do meaningful reasoning, it has glimpses of intuition. When given observations from past landmark experiments, the model can declare that “light is made up of definite quantities of energy” and even suggest that gravity and acceleration are locally equivalent. I’m releasing the dataset + models and leave this as an open problem to the research community. I also include what this project has taught me about intelligence in a mini essay linked below. 🧵(1/n)
→ View original post on X — @thom_wolf, 2026-04-02 18:14 UTC
-
Google Releases Gemma 4: Open-Source Agentic Model Breakthrough
By
–
Google just released Gemma 4, and from an outside perspective, it is a massive leap for the open-source community. They have successfully packaged Gemini 3 technology into a highly capable, fully open agentic model. What is genuinely impressive is the licensing. It ships
-
x402 Protocol Joins Linux Foundation as Payment Primitive
By
–
ICYMI: x402 just landed a home under the Linux Foundation The same neutral nonprofit that stewards Linux, Kubernetes and some of the most critical open source infrastructure on the planet when a protocol lands here it is a signal that the industry considers it foundational TLDR: i) Coinbase, Cloudflare, Stripe, Google, Microsoft, AWS, Visa, Mastercard, Amex, Shopify, Circle, Solana Foundation and others are all founding members of the x402 Foundation ii) x402 is the payment primitive the web never had; HTTP moves data, x402 moves money, natively at the protocol level iii) rail-agnostic; fiat, crypto, multi-chain, no ideological gatekeeping iv) built for the agentic web; agents need to transact programmatically without human authorization loops v) no single company owns it; that is the point we published a full @KhalaResearch deep dive on x402 last month covering the architecture, facilitator model and the broader agentic commerce stack (visual below) I'll link the full report below: Coinbase 🛡️ (@coinbase) x.com/i/article/203965656653… — https://nitter.net/coinbase/status/2039689438922522728#m
→ View original post on X — @whiteafrican, 2026-04-02 17:27 UTC
-
NVIDIA Releases Quantized Gemma 4 31B Model on Hugging Face
By
–
NVIDIA just released a quantized Gemma 4 31B on Hugging Face NVFP4 compression delivers 4x smaller weights with frontier-level accuracy. Runs on consumer GPUs with a 256K context window.
→ View original post on X — @huggingface, 2026-04-02 17:17 UTC
-
llama.cpp achieves 300 tokens/second on Mac Studio M2 Ultra
By
–
Let me demonstrate the true power of llama.cpp:
— Georgi Gerganov (@ggerganov) 2 avril 2026
– Running on Mac Studio M2 Ultra (3 years old)
– Gemma 4 26B A4B Q8_0 (full quality)
– Built-in WebUI (ships with llama.cpp)
– MCP support out of the box (web-search, HF, github, etc.)
– Prompt speculative decoding
The result:… pic.twitter.com/B3EnpbWJdeLet me demonstrate the true power of llama.cpp: – Running on Mac Studio M2 Ultra (3 years old) – Gemma 4 26B A4B Q8_0 (full quality) – Built-in WebUI (ships with llama.cpp) – MCP support out of the box (web-search, HF, github, etc.) – Prompt speculative decoding The result: 300t/s (realtime video)
-
Google releases Apache 2.0 open-source model
By
–
The world wants Apache 2.0 & open-source! Thank you @Google @demishassabis @sundarpichai