Papersby GlyphGemma 4 Technical Report research paper by Google DeepMind, 2026
Google DeepMind · Foundation models · Jul 2, 2026 · 356 citations · 81 upvotes
What it shows
Open multimodal models from 2.3B to 31B parameters, with a thinking mode and image and audio input.
By Gemma Team, Sherif El Abd, Vaibhav Aggarwal and 298 more · arXiv 2607.02770 · PDF
Summarised by hand from the abstract.
Models it introduced
On models.fru.devMore from Google DeepMind
All 12DiffusionGemma Technical ReportDiffusionGemma is a fine-tuned mixture-of-experts language model that uses discrete diffusion to generate text blocks in parallel, achieving high speed while preserving capabilities like multimodal inputs and reasoning.
Google DeepMind14 citations40 upvotesRead →LiteFrame: Efficient Vision Encoders Unlock Frame Scaling in Video LLMsLiteFrame, a lightweight video encoder with Compressed Token Distillation training method, reduces latency and increases frame processing capacity for long-form video understanding in Video LLMs while maintaining accuracy.Google DeepMind2 citations26 upvotesRead →Understanding the Challenges in Iterative Generative Optimization with LLMsGenerative optimization using large language models faces challenges due to implicit design decisions about artifact modification and learning evidence that significantly impact success across different applications.Google DeepMind11 citations27 upvotesRead →LoGeR: Long-Context Geometric Reconstruction with Hybrid MemoryLoGeR enables long-term 3D video reconstruction by combining bidirectional priors with a hybrid memory system that includes parametric Test-Time Training and non-parametric sliding window attention mechanisms.Google DeepMind46 citations63 upvotesRead →SIMA 2: A Generalist Embodied Agent for Virtual WorldsSIMA 2, built on a Gemini foundation model, interacts in 3D virtual worlds, reasons about goals, handles complex instructions, and autonomously learns new skills through open-ended self-improvement.Google DeepMind27 citations26 upvotesRead →Robot Learning from a Physical World ModelPhysWorld integrates video generation and physical world modeling to enable accurate robotic manipulation from visual demonstrations without real robot data.Google DeepMind23 citations32 upvotesRead →Vibe Checker: Aligning Code Evaluation with Human PreferenceVibe Checker evaluates LLMs by combining functional correctness and instruction following to better align with human coding preferences.Google DeepMind2 citations34 upvotesRead →Video models are zero-shot learners and reasonersVeo 3, a generative video model, exhibits zero-shot capabilities across various visual tasks, suggesting a trajectory towards becoming a unified, generalist vision foundation model.Google DeepMind237 citations100 upvotesRead →
Releases that name it
From releases.fru.devImport Google Gemma 4 26B A4B IT into OCI Generative AIoraclereleases.fru.dev →Gemma-4-31B-it-assistant and Gemma-4-31B-IT-NVFP4 models now available on Amazon SageMaker JumpStartawsreleases.fru.dev →Import Google Gemma 4 31B IT into OCI Generative AIoraclereleases.fru.dev →Gemma 4 31B availability changes on public endpoints Starting September 3, 2026, gemma-4-31b is no longer...cerebrasreleases.fru.dev →Gemma 4 31B available in AIP through AWS Bedrockpalantirreleases.fru.dev →We now support prompt caching for Gemma 4 for serverless inferencedigitaloceanreleases.fru.dev →Gemma 4 31B available across coding integrations gemma-4-31b is now a selectable model in our Cline, Kilo...cerebrasreleases.fru.dev →Gemma 4 31B and Image Input Support - Gemma 4 31B (gemma-4-31b) is now available in previewcerebrasreleases.fru.dev →Workers AI - Google Gemma 4 26B A4B now available on Workers AIcloudflarereleases.fru.dev →
Other foundation models papers
TopicBERT: Pre-training of Deep Bidirectional Transformers for Language UnderstandingPre-training a Transformer to fill in masked words, then fine-tuning it, beat task-specific models across language understanding tests.
Google125k citations33 upvotesRead →Language Models are Few-Shot LearnersGPT-3, a 175B-parameter model, does new tasks from a few examples in the prompt, with no fine-tuning.
OpenAI66.4k citations21 upvotesRead →GPT-4 Technical ReportA multimodal model that passes a simulated bar exam with a score around the top 10% of test takers.OpenAI28.4k citations7 upvotesRead →LLaMA: Open and Efficient Foundation Language ModelsOpen models from 7B to 65B trained only on public data; the 13B model beats GPT-3 on most benchmarks.
Meta22.6k citations25 upvotesRead →The Llama 3 Herd of ModelsOpen models up to 405B parameters that match leading closed models on many tasks.Meta20.8k citations119 upvotesRead →Qwen3 Technical ReportQwen3, a unified series of large language models, integrates thinking and non-thinking modes, reduces computational resources, and achieves state-of-the-art performance across various tasks and languages.
Alibaba (Qwen)10.1k citations339 upvotesRead →
About this paper
- Venue
- arXiv.org
- Citations
- 356, 48 influential · Semantic Scholar
- Upvotes
- 81 · Hugging Face
- Lab
- Google DeepMind · on Companies · on Acquisitions
Changes
Influential citations rose from 43 to 48.
Citations rose from 321 to 356.
Influential citations rose from 28 to 43.
Citations rose from 220 to 321.
Influential citations rose from 23 to 28.
Citations rose from 181 to 220.
Influential citations rose from 14 to 23.
Citations rose from 117 to 181.
Influential citations rose from 13 to 14.
Citations rose from 110 to 117.
Influential citations rose from 12 to 13.
Citations rose from 106 to 110.
Added to the list.