Kaggle (Organization): news timeline & CVEs

EmbeddingGemma 2: an open, lightweight multimodal embedding model

Google DeepMind released EmbeddingGemma 2, a 740-million-parameter open multimodal embedding model for on-device use.

Google DeepMind released EmbeddingGemma 2, a 740-million-parameter Apache 2.0 model built on Gemma 4 that maps text, code, images, video, and audio into one embedding space. It reports a 9.92-point MTEB Code gain, from 68.76 to 78.68, and can truncate vectors from 768 to 128 dimensions via Matryoshka Representation Learning. Text-only workloads can use about 270 million parameters, with optional 170-million-parameter vision and 300-million-parameter audio encoders. On a Pixel 11 Pro, quantized weights need about 191 MB RAM for text and 567 MB for the full model; context is 8K tokens. Weights are on Hugging Face and Kaggle.

Google DeepMindOct 6, 19:57 UTC · 2d agoModel release 4 sources

Google claims EmbeddingGemma 2 outperforms rival embedding models twice its size

Google released EmbeddingGemma 2, a 740-million-parameter open multimodal embedding model.

Google released EmbeddingGemma 2, an open model that turns text, images, video, audio, and code into vectors for similarity search. At 740 million parameters, Google says it is the most compact model of its kind and outperforms rivals up to twice its size on multimodal embedding benchmarks. It scores 78.68 on MTEB Code, up from 68.76, runs locally in about 20 to 70 milliseconds, and uses roughly 191 MB of RAM. A 270-million-parameter text-only variant is available, with weights on Hugging Face and Kaggle.

The DecoderOct 6, 19:47 UTC · 2d agoModel release 4 sources

EmbeddingGemma 2: An open, lightweight multimodal embedding model

Google DeepMind released EmbeddingGemma 2, a 740M open multimodal embedding model for on-device search.

Google DeepMind released EmbeddingGemma 2, a 740-million-parameter open multimodal embedding model under Apache 2.0, built on the Gemma 4 architecture. It maps text, code, images, audio, and video into one space, with an 8K context window and Matryoshka vectors from 768 down to 128 dimensions. On MTEB Code it scores 78.68, a 9.92-point gain over EmbeddingGemma, and leads sub-1B multimodal embedders on several benchmarks. Quantized weights use about 191MB RAM for text-only and about 567MB for the full model on a Pixel 11 Pro. Weights are available on Hugging Face and Kaggle.

Hacker News · AIupdated Oct 6, 19:57 UTC · 2d agofirst Oct 6, 16:03 UTC · 2d agoModel release 4 sources HN 156↑ · 21 comments

Game Arena: Strategic LLM Evaluation in Competitive Environments

Kaggle Game Arena ranks LLMs through ongoing Chess, Poker, and Werewolf competitions instead of static benchmarks.

The report introduces Kaggle Game Arena, an open platform for evaluating LLMs through competitive games rather than static benchmarks. Its three pilot environments are Chess, Poker, and Werewolf, covering perfect-information, imperfect-information, and multiplayer settings. For each game it describes the environment, metrics, and results from full model competitions. The infrastructure is designed for reproducible, ground-truth evaluation and for adding new games over time.

Hugging Face daily papersSep 24, 20:00 UTC · 14d agoAI research

Kaggle

Timeline4 articles

Appears with

EmbeddingGemma 2: an open, lightweight multimodal embedding model

Google claims EmbeddingGemma 2 outperforms rival embedding models twice its size

EmbeddingGemma 2: An open, lightweight multimodal embedding model

Game Arena: Strategic LLM Evaluation in Competitive Environments