FOLLOW A THREAD

多模态模型

Connect scattered updates into a story you can understand.

ProjectsNews10/17, 10:59

Perplexity open-sources two decision models: per million inputs Token as low as 1 cents, with inference speed up to 5 times

Perplexity open-sources 27B and 4B two multimodal decision models that support classification, scoring, and judgment tasks. The official statement says inference speed improves by up to 5 times in some scenarios, while API input prices start at per million Token 0.01 USD。

MarketUnverified10/08, 19:35

Scale Labs Benchmark finds 25 multimodal models lag behind humans in common-sense understanding

Beating AI reports that Scale Labs and Elorian released the HSS benchmark, and that 25 multimodal models all scored below human participants in tests of understanding implicit information in images and videos. The report also states that increasing reasoning effort and using image-inspection Agents helped, but did not close the gap.

# Industry Insights # Multimodal models # AI Benchmark # Computer vision # Common-sense understanding

ProjectsUnverified10/07, 11:00

Google DeepMind open-sources EmbeddingGemma 2, supporting unified retrieval of text, images, video, and audio

Google DeepMind open-sources EmbeddingGemma 2, supporting unified multimodal vector retrieval, with up to 740000000 parameters and using the Apache 2.0 license.

# Technology Developments # multimodal model # on-device AI # vector retrieval # RAG

ProjectsNews09/30, 19:19

SGLang Adds native decision-making API, planned for official release with v0.5.21

According to reports,SGLang added native decision-making in its nightly version API and the /v1/systemone endpoint, enabling off-the-shelf LLM to directly return the probabilities of candidate options. The feature is planned for official release with v0.5.21.

# Industry Insights # SGLang # Large-model inference # Decision-making API # Multimodal model

All loaded4 shown · 12 per batch