--- title: Google Updates Gemma 4: The best Small LLM is back !! description: Google has released a significant update to Gemma 4, its open-weight multimodal model family, without changing the underlying model weights. The update... canonical: https://daily.dev/posts/google-updates-gemma-4-the-best-small-llm-is-back--tvolamyq7 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Google Updates Gemma 4: The best Small LLM is back !! | daily.dev og:description: Google has released a significant update to Gemma 4, its open-weight multimodal model family, without changing the underlying model weights. The update... og:url: https://daily.dev/posts/google-updates-gemma-4-the-best-small-llm-is-back--tvolamyq7 og:image: https://api.daily.dev/og/posts/TvOLAMyq7.png og:image:alt: Google Updates Gemma 4: The best Small LLM is back !! og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Google Updates Gemma 4: The best Small LLM is back !! **[Medium](https://daily.dev/sources/medium_js)** · 6 min read · 1 upvotes · 0 comments ## Summary Google has released a significant update to Gemma 4, its open-weight multimodal model family, without changing the underlying model weights. The update addresses five key areas: reliable tool calling (fixing malformed outputs in long conversations), a configurable vision token budget for balancing inference cost vs. image detail, Flash Attention 4 support for NVIDIA Hopper GPUs (H100/H200/GH200), improved chat templates for better prompt formatting and compatibility, and reduced model laziness for more complete responses. Benchmark results show 8-point improvement in airline tool-calling tasks and 6+ points in telecom tasks. Developers using vLLM can immediately benefit from updated parsers and chat templates. The update is especially valuable for teams building AI agents, document processing pipelines, and OCR systems. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What changed in the latest Gemma 4 update if the model weights stayed the same? Google improved five capabilities without shipping new weights: more reliable tool calling, a configurable vision token budget, Flash Attention 4 support for NVIDIA Hopper GPUs, an improved chat template, and reduced model laziness in responses. Previously Gemma 4 sometimes returned malformed tool outputs as plain text instead of structured function calls; this is now fixed and confirmed working in vLLM deployments. _Teams running agent workloads on gemma track updates like this on daily.dev before touching production configs._ ### Does Flash Attention 4 support in Gemma 4 help if I'm running on older GPUs? No, Flash Attention 4 support is designed specifically for NVIDIA Hopper GPUs, including the H100, H200, and GH200. It delivers faster inference, better throughput, improved GPU utilization, and lower latency on that hardware, but the optimization does not apply to older NVIDIA architectures or consumer GPUs. _Developers weighing gpu upgrades for llm inference can follow hardware-specific updates like this on daily.dev._ ### How much did Gemma 4's tool calling accuracy improve according to Google's benchmarks? Google's internal benchmarks show airline-related tool calling tasks improved by approximately 8 points and telecom tasks improved by more than 6 points after the update. These gains come from fixes to how the model generates structured function calls instead of returning malformed plain text during long conversations, verified in vLLM testing. _Anyone benchmarking agent reliability with gemma can keep tabs on model updates like this via daily.dev._ ## Similar posts on daily.dev - [Google’s Gemma 4 shines on local systems – both big and small](https://daily.dev/posts/google-s-gemma-4-shines-on-local-systems-both-big-and-small-qpert58dk) · InfoWorld · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#multimodal](https://daily.dev/tags/multimodal), [#vllm](https://daily.dev/tags/vllm), [#gemma](https://daily.dev/tags/gemma) [View this post on daily.dev](https://daily.dev/posts/google-updates-gemma-4-the-best-small-llm-is-back--tvolamyq7) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Google Updates Gemma 4: The best Small LLM is back !!","url":"https://daily.dev/posts/google-updates-gemma-4-the-best-small-llm-is-back--tvolamyq7","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/google-updates-gemma-4-the-best-small-llm-is-back--tvolamyq7"},"datePublished":"2026-07-23T05:10:51.451Z","dateModified":"2026-09-14T07:15:04.345Z","description":"Google has released a significant update to Gemma 4, its open-weight multimodal model family, without changing the underlying model weights. The update...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/e79b488214fbb4f9c5494398bc48d5f5?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/e79b488214fbb4f9c5494398bc48d5f5?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Medium","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Medium","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/medium","url":"https://daily.dev/sources/medium_js"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/google-updates-gemma-4-the-best-small-llm-is-back--tvolamyq7","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,multimodal,vllm,gemma","timeRequired":"PT6M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Medium","item":"https://daily.dev/sources/medium_js"},{"@type":"ListItem","position":3,"name":"Google Updates Gemma 4: The best Small LLM is back !!"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/google-updates-gemma-4-the-best-small-llm-is-back--tvolamyq7#faq","mainEntity":[{"@type":"Question","name":"What changed in the latest Gemma 4 update if the model weights stayed the same?","acceptedAnswer":{"@type":"Answer","text":"Google improved five capabilities without shipping new weights: more reliable tool calling, a configurable vision token budget, Flash Attention 4 support for NVIDIA Hopper GPUs, an improved chat template, and reduced model laziness in responses. Previously Gemma 4 sometimes returned malformed tool outputs as plain text instead of structured function calls; this is now fixed and confirmed working in vLLM deployments. Teams running agent workloads on gemma track updates like this on daily.dev before touching production configs."}},{"@type":"Question","name":"Does Flash Attention 4 support in Gemma 4 help if I'm running on older GPUs?","acceptedAnswer":{"@type":"Answer","text":"No, Flash Attention 4 support is designed specifically for NVIDIA Hopper GPUs, including the H100, H200, and GH200. It delivers faster inference, better throughput, improved GPU utilization, and lower latency on that hardware, but the optimization does not apply to older NVIDIA architectures or consumer GPUs. Developers weighing gpu upgrades for llm inference can follow hardware-specific updates like this on daily.dev."}},{"@type":"Question","name":"How much did Gemma 4's tool calling accuracy improve according to Google's benchmarks?","acceptedAnswer":{"@type":"Answer","text":"Google's internal benchmarks show airline-related tool calling tasks improved by approximately 8 points and telecom tasks improved by more than 6 points after the update. These gains come from fixes to how the model generates structured function calls instead of returning malformed plain text during long conversations, verified in vLLM testing. Anyone benchmarking agent reliability with gemma can keep tabs on model updates like this via daily.dev."}}]} ```