Gemma Scope 2 is a comprehensive, open suite of sparse autoencoders and transcoders for a range of model sizes and versions in the Gemma 3 model family.
Running this yourself: consumer gpu should be enough.
Model updates refreshed5d agoOct 31, 2026news + changelog
Live access is not confirmed for this model. No current purchase price is advertised.
Related provider subscriptions
No current subscription pricing is tracked for this model.
Confirm this specific model, usage limits, and billing terms with the provider. A subscription does not automatically include API credits.
No login needed to compare. Prices are in USD; provider charges are separate from AI Market Cap plans. Context length, caching, tools, taxes, and regional terms can change the final cost. Open weights do not mean free hosting.
20.3
Quality Score
---
Arena ELO
4B
Parameters
---
Context
Evidence profile
How complete is this record?
This measures the amount of verifiable public evidence we have, not how capable the model is. A missing field means it has not been verified yet, not that its value is zero.
Recent launch, pricing, benchmark, and API signals linked to this model or its provider.
LaunchesGoogle1w ago
Introducing Genie 4 from @GoogleDeepMind, our most realistic and interactive world model yet 🌍 Now you can use natural language to dynamically steer the world and its soundscape in real-time, whether
Introducing Genie 4 from @GoogleDeepMind, our most realistic and interactive world model yet 🌍 Now you can use natural language to dynamically steer the world and its soundscape in real-time, whether you're changing the weather, moving the camera, or introducing new elements to https://t.co/ZfZGi9Chsx
It’s Friday 🎉 Here’s a look at this week’s announcements: — https://t.co/a9bSlm0Yxk is now available globally in English, allowing anyone to use our detector portal to easily verify if images, video,
It’s Friday 🎉 Here’s a look at this week’s announcements: — https://t.co/a9bSlm0Yxk is now available globally in English, allowing anyone to use our detector portal to easily verify if images, video, or audio files were AI-generated by Google or industry partners — Nano Banana
Quaternion Approximate Networks for Enhanced Image Classification and Oriented Object Detection
This paper introduces Quaternion Approximate Networks (QUAN), a novel deep learning framework that leverages quaternion algebra for rotation equivariant image classification and object detection. Unlike conventional quaternion neural networks attempting to operate entirely in the quaternion domain, QUAN approximates quaternion convolution through Hamilton product decomposition using real-valued operations. This approach preserves geometric properties while enabling efficient implementation with custom CUDA kernels. We introduce Independent Quaternion Batch Normalization (IQBN) for training stability and extend quaternion operations to spatial attention mechanisms. QUAN is evaluated on image classification (CIFAR-10/100, ImageNet), object detection (COCO, DOTA), and robotic perception tasks. In classification tasks, QUAN achieves higher accuracy with fewer parameters and faster convergence compared to existing convolution and quaternion-based models. For objection detection, QUAN demonstrates improved parameter efficiency and rotation handling over standard Convolutional Neural Networks (CNNs) while establishing the SOTA for quaternion CNNs in this downstream task. These results highlight its potential for deployment in resource-constrained robotic systems requiring rotation-aware perception and application in other domains. Code: https://cwru-aism.github.io/ QUANpaper/
EvaOmni: An Evidence-Grounded Benchmark for Audio-Visual Reasoning
Omni LLMs can naturally locate key audio and visual evidence and integrate information across modalities for audiovisual reasoning. Existing audio-visual reasoning benchmarks that mainly rely on multiple-choice questions usually score models solely based on answer correctness, making it difficult to assess whether a model truly forms a reasoning chain from evidence to the correct answer. In addition, textual or unimodal shortcuts may allow models to answer correctly without analyzing the necessary audio-visual evidence, further weakening the reliability of evaluation benchmarks. To address these issues, we introduce EvaOmni, an evidence-grounded benchmark containing 392 videos and 1,027 purely human-annotated question-answer pairs, covering 15 capabilities and three levels: retrieval and localization, basic understanding, and complex reasoning. We design a series of rigorous annotation procedures in which experts annotate audio and visual evidence and construct video-based distractors; humans review whether questions meet strict inspection criteria, and finally models filter textual and unimodal shortcuts. Beyond high-quality question-answer pairs, we require models to simultaneously return the answer and supporting evidence intervals in order to evaluate the complete reasoning chain. We propose the Coverage Grounded Score (C.G.) to measure how much annotated evidence a model identifies, and the Bottleneck Grounded Score (B.G.) to penalize poor temporal localization in either modality; both are incorporated into answer correctness. Across 7 open-weight and 17 proprietary Omni LLMs, Gemini3.7-Flash achieved the highest overall accuracy (79.4%), C.G. (70.8%), and B.G. (35.1%). At Level 3, it reached 79.1% accuracy, but its B.G. was only 12.3%. These results show that even the most advanced models still have room for development in grounded audio-visual reasoning.
Diffusion Language Models Change Speed Across Tasks and Languages
Diffusion language models can write several tokens in one forward pass, and they decide during generation how many passes to spend. Two outputs of the same token length can therefore vary a lot in cost. To understand the underlying factors of this cost variation, we analyze two diffusion models across 10 tasks. We find that the number of forward passes per token depends heavily on the task. Text copied or lightly edited from the input is cheap, while open-ended writing is costly. In DiffusionGemma, generating a token for story writing requires more than three times as many forward passes as for grammar correction. Two output properties explain much of this gap: how much of it is new relative to the input, and how rare its words are within the task. The same holds within a single response when thinking is involved. The final answer is cheap when it repeats the model's reasoning trace and costly when it adds something new. In multilingual generation, we find that highly tokenized languages also require more passes per token, adding to the cost they already pay with more tokens. Finally, cost tracks correctness: on the same math problem, wrong answers take more passes per token than correct ones, and weighting candidate answers by decoding speed beats majority voting by up to 4.2 points at no extra cost. Taken together, our results show that diffusion's computational advantage depends on how much of the output is already determined, while its decoding effort offers a practical signal for making inference more efficient and reliable.
We built PlanAssist to help UK Government planning officers speed up building and renovations across the country. Built with Gemini, it has shown it can cut the time planning officers spend assessing
We built PlanAssist to help UK Government planning officers speed up building and renovations across the country. Built with Gemini, it has shown it can cut the time planning officers spend assessing householder applications by up to 58%. It is now being made available to 12 https://t.co/qab3z41WDa
Introducing Genie 4 from @GoogleDeepMind, our most realistic and interactive world model yet 🌍 Now you can use natural language to dynamically steer the world and its soundscape in real-time, whether
Introducing Genie 4 from @GoogleDeepMind, our most realistic and interactive world model yet 🌍 Now you can use natural language to dynamically steer the world and its soundscape in real-time, whether you're changing the weather, moving the camera, or introducing new elements to https://t.co/ZfZGi9Chsx
X/Twitter@GoogleDeepMindGoogleannouncementgeneral1w ago
This is Genie 4, our most realistic and interactive world model yet. 🌐 Explore dynamic environments in real time to see how the scene responds, sounds, and how the people in it move, react, and speak
This is Genie 4, our most realistic and interactive world model yet. 🌐 Explore dynamic environments in real time to see how the scene responds, sounds, and how the people in it move, react, and speak. https://t.co/6mmSUm9BBO
X/Twitter@GoogleDeepMindGoogleannouncementgeneral3w ago
Over half of reported crimes in Singapore use sophisticated social engineering. With @GovTechSG, we’re building on-device fraud detection using our open model Gemma. By fine-tuning it on local scam ta
Over half of reported crimes in Singapore use sophisticated social engineering. With @GovTechSG, we’re building on-device fraud detection using our open model Gemma. By fine-tuning it on local scam tactics, it identifies threats in real time without compromising personal https://t.co/pkAulZoJng
It’s Friday 🎉 Here’s a look at this week’s announcements: — https://t.co/a9bSlm0Yxk is now available globally in English, allowing anyone to use our detector portal to easily verify if images, video,
It’s Friday 🎉 Here’s a look at this week’s announcements: — https://t.co/a9bSlm0Yxk is now available globally in English, allowing anyone to use our detector portal to easily verify if images, video, or audio files were AI-generated by Google or industry partners — Nano Banana
Quaternion Approximate Networks for Enhanced Image Classification and Oriented Object Detection
This paper introduces Quaternion Approximate Networks (QUAN), a novel deep learning framework that leverages quaternion algebra for rotation equivariant image classification and object detection. Unlike conventional quaternion neural networks attempting to operate entirely in the quaternion domain, QUAN approximates quaternion convolution through Hamilton product decomposition using real-valued operations. This approach preserves geometric properties while enabling efficient implementation with custom CUDA kernels. We introduce Independent Quaternion Batch Normalization (IQBN) for training stability and extend quaternion operations to spatial attention mechanisms. QUAN is evaluated on image classification (CIFAR-10/100, ImageNet), object detection (COCO, DOTA), and robotic perception tasks. In classification tasks, QUAN achieves higher accuracy with fewer parameters and faster convergence compared to existing convolution and quaternion-based models. For objection detection, QUAN demonstrates improved parameter efficiency and rotation handling over standard Convolutional Neural Networks (CNNs) while establishing the SOTA for quaternion CNNs in this downstream task. These results highlight its potential for deployment in resource-constrained robotic systems requiring rotation-aware perception and application in other domains. Code: https://cwru-aism.github.io/ QUANpaper/
EvaOmni: An Evidence-Grounded Benchmark for Audio-Visual Reasoning
Omni LLMs can naturally locate key audio and visual evidence and integrate information across modalities for audiovisual reasoning. Existing audio-visual reasoning benchmarks that mainly rely on multiple-choice questions usually score models solely based on answer correctness, making it difficult to assess whether a model truly forms a reasoning chain from evidence to the correct answer. In addition, textual or unimodal shortcuts may allow models to answer correctly without analyzing the necessary audio-visual evidence, further weakening the reliability of evaluation benchmarks. To address these issues, we introduce EvaOmni, an evidence-grounded benchmark containing 392 videos and 1,027 purely human-annotated question-answer pairs, covering 15 capabilities and three levels: retrieval and localization, basic understanding, and complex reasoning. We design a series of rigorous annotation procedures in which experts annotate audio and visual evidence and construct video-based distractors; humans review whether questions meet strict inspection criteria, and finally models filter textual and unimodal shortcuts. Beyond high-quality question-answer pairs, we require models to simultaneously return the answer and supporting evidence intervals in order to evaluate the complete reasoning chain. We propose the Coverage Grounded Score (C.G.) to measure how much annotated evidence a model identifies, and the Bottleneck Grounded Score (B.G.) to penalize poor temporal localization in either modality; both are incorporated into answer correctness. Across 7 open-weight and 17 proprietary Omni LLMs, Gemini3.7-Flash achieved the highest overall accuracy (79.4%), C.G. (70.8%), and B.G. (35.1%). At Level 3, it reached 79.1% accuracy, but its B.G. was only 12.3%. These results show that even the most advanced models still have room for development in grounded audio-visual reasoning.
Diffusion Language Models Change Speed Across Tasks and Languages
Diffusion language models can write several tokens in one forward pass, and they decide during generation how many passes to spend. Two outputs of the same token length can therefore vary a lot in cost. To understand the underlying factors of this cost variation, we analyze two diffusion models across 10 tasks. We find that the number of forward passes per token depends heavily on the task. Text copied or lightly edited from the input is cheap, while open-ended writing is costly. In DiffusionGemma, generating a token for story writing requires more than three times as many forward passes as for grammar correction. Two output properties explain much of this gap: how much of it is new relative to the input, and how rare its words are within the task. The same holds within a single response when thinking is involved. The final answer is cheap when it repeats the model's reasoning trace and costly when it adds something new. In multilingual generation, we find that highly tokenized languages also require more passes per token, adding to the cost they already pay with more tokens. Finally, cost tracks correctness: on the same math problem, wrong answers take more passes per token than correct ones, and weighting candidate answers by decoding speed beats majority voting by up to 4.2 points at no extra cost. Taken together, our results show that diffusion's computational advantage depends on how much of the output is already determined, while its decoding effort offers a practical signal for making inference more efficient and reliable.
From Failures to Learnable Worlds: Policy-Conditioned Environment Generation for Navigation Agents
Simulated environments enable scalable training data generation for embodied agents, but more data does not necessarily provide what a particular agent needs to learn. Existing environment-generation pipelines largely optimize generic properties such as realism, diversity, or task coverage, while different pretrained policies can fail for different environmental reasons. We therefore introduce WEDGE, a framework for weakness-guided navigation learning in which the failure modes of a pretrained agent guide the generation of training environments at each adaptation round. Given a pretrained policy, WEDGE uses an LLM to propose executable weakness features from contrasting successful and failed rollouts, automatically validates them on the diagnostic set, and selects the strongest validated weakness based on its success-rate gap. The validated weaknesses are then mapped to the environmental factors that control them. WEDGE varies these factors to construct environments in the difficulty regimes where the policy struggles, producing targeted training experience for adaptation. Across three navigation agents with different adaptation mechanisms, we find that policies exhibit distinct weakness profiles and benefit from different training environments. Weakness-conditioned adaptation consistently outperforms matched random and hard-example alternatives on ProcTHOR, while the resulting gains for Uni-NaVid and Gemini also extend to HM3D after ProcTHOR adaptation. Iterative re-diagnosis further shows that dominant weaknesses can persist or shift as the policy improves, providing updated guidance for subsequent environment construction. Together, these results show that policy failures are not only diagnostic signals, but actionable guidance for constructing training environments that target a policy's current weaknesses.
APEX: Asynchronous Prefix Denoising for Extreme Compression
Fewer latent tokens promise cheaper diffusion. Yet aggressive latent compression creates a reconstruction-generation dilemma: adding channels improves reconstruction but makes generation harder. We introduce APEX, short for Asynchronous Prefix denoising for EXtreme compression, which pairs a hierarchical latent space with asynchronous denoising over channel groups. Our autoencoder orders channel prefixes from structure to detail, while one shared denoiser generates them at staggered local times without adding tokens. On ImageNet-512, APEX outperforms existing methods at every evaluated channel count under 64x and 128x compression. At matched training and guided sampling compute, it also outperforms a 256-token model at every evaluated budget, with larger gains at lower compute. Code is available at https://daviddinkevich.github.io/APEX.
