# Gemma 4 713 Hacker News commenters recommended Gemma 4 and 249 criticized it, November 2025 to November 2026. It was recommended in 13 of 13 months; owners who said how long report a median of 0.1 years (14 owners). It meets the Hacker Approved badge rule. **Verdict:** Google's open-weight model family runs well on local hardware with strong quality and token efficiency for its size, though critics cite unreliable tool calling and weaker coding than Qwen. ## Why people recommend it - Runs well on local hardware (8 comments): "the release of open-weight Gemma models that can run on reasonable local hardware, and are not drastically worse than Anthropic flagships, is quite a punch" ([source](https://news.ycombinator.com/item?id=47771651)) "which you can run on an off-the-shelf laptop, is around the same or even higher intelligence level as the SOTA models from 2 years ago" ([source](https://news.ycombinator.com/item?id=47823052)) "The hardware for 50 tokens per second with a four bit quantisation of Gemma 4 26B or the sparse Qwen 3.6 is not really that expensive" ([source](https://news.ycombinator.com/item?id=48425284)) - Best-in-class quality for size (6 comments): "I have banged on about this model here enough but I really agree that Gemma 4 12B is a candidate for the most impressive LLM of the year." ([source](https://news.ycombinator.com/item?id=49330061)) "Qwen 3.8 27B is only the second local model after Gemma 4 that managed to correctly reason through one of my private benchmarks." ([source](https://news.ycombinator.com/item?id=49310350)) "better than anything you can run locally in ~8GB" ([source](https://news.ycombinator.com/item?id=48741184)) - Fast inference and token-efficient (6 comments): "there are complex tasks that gemma will complete much much faster than qwen because it makes the right assumptions the first time" ([source](https://news.ycombinator.com/item?id=47966020)) "Impressive speed and quality for the small model with thinking off (E4B). Very slow for the big model with thinking turned on." ([source](https://news.ycombinator.com/item?id=47706098)) "use enormously less tokens per output than other models, while still staying within arms reach of top benchmark performance" ([source](https://news.ycombinator.com/item?id=48026531)) - Strong vision and general tasks (5 comments): "better than anything you can run locally in ~8GB" ([source](https://news.ycombinator.com/item?id=48741184)) "Gemma 4 26B outperforms most 100-200B models that I've tested for reasoning and structured output." ([source](https://news.ycombinator.com/item?id=48417879)) "understands a lot of my references and influences, could recommend me my (still semi-obscure) favourite historical photographer" ([source](https://news.ycombinator.com/item?id=49254713)) ## What people warn about - Unreliable tool calling (5 comments): "The model does call tools successfully giving sensible parameters but it seems to not picking the right ones in the right order." ([source](https://news.ycombinator.com/item?id=47619810)) "I did get one weird session where the model got confused and couldn't decide which tools actually existed in its inventory" ([source](https://news.ycombinator.com/item?id=47839640)) "Reminds me of Gemma4 and the official (or at least popularly used around launch) Jinja templates being wrong and broken for tool calling." ([source](https://news.ycombinator.com/item?id=49327465)) - Qwen better for coding/agents (3 comments): "there are tasks where Gemma is great, but on my camera watching, Qwen 3.6's MoE far exceeds either the Gemma 4 Dense or MoE" ([source](https://news.ycombinator.com/item?id=49291388)) "Gemma 4 31b is a good model for other tasks but at least my personal experience is that Qwen outperforms in coding." ([source](https://news.ycombinator.com/item?id=48544883)) "better than anything you can run locally in ~8GB" ([source](https://news.ycombinator.com/item?id=48741184)) - Mixes in foreign-script characters (2 comments): "it produced mostly good code for simple problems, but it sometimes hallucinated words in non-English scripts inside the code" ([source](https://news.ycombinator.com/item?id=47636746)) "it loved to just stick Chinese and Korean characters into the text, sometimes without seeming to realize it had mixed languages" ([source](https://news.ycombinator.com/item?id=50187765)) - Smallest variants are weak (3 comments): "even "old" models like llama 3.2, gemma 2, qwen 2.5, are good, fast and low-memory" ([source](https://news.ycombinator.com/item?id=48095563)) "That's a tiny model. No LLM is going to be a perfect and repeatable judge, but a tiny 4B model is like plugging an RNG into this system." ([source](https://news.ycombinator.com/item?id=48715422)) "They fail at even the most basic questions I throw at them, like 12B just now failed at answering how XGrabKey from Xlib is used." ([source](https://news.ycombinator.com/item?id=49343186)) ## Alternatives hackers recommend - [Qwen 3.8 27B](https://hackerapproved.com/p/qwen-3-8-27b.md): 1173 recommend - [Z.ai GLM](https://hackerapproved.com/p/glm-5-2.md): 1043 recommend - [Kimi](https://hackerapproved.com/p/kimi-k3.md): 931 recommend - [Qwen](https://hackerapproved.com/p/qwen.md): 474 recommend - [Qwen3.6-35B-A3B](https://hackerapproved.com/p/qwen3-6-35b-a3b.md): 373 recommend ## Cite as Hacker Approved, "Gemma 4", https://hackerapproved.com/p/gemma-4/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).