Google releases multimodal Gemma 4-12B-it model

// 1h agoMODEL RELEASE

Google releases multimodal Gemma 4-12B-it model

Google has announced the release of Gemma 4-12B-it, a new open multimodal model that is part of the broader Gemma 4 family. The model offers significant advancements in natively handling and generating various types of inputs and outputs, facilitating easier integration of advanced AI capabilities by developers.

// ANALYSIS

Google's new 12B open-weights model is a major step forward for local development, combining efficiency with native multimodal support.

–Gemma 4-12B-it runs efficiently on local hardware with approximately 16GB of VRAM.
–The encoder-free architecture integrates vision and audio natively into the LLM backbone, improving latency and performance.
–Released under the Apache 2.0 license, it encourages open ecosystem adoption and integration.

// TAGS

gemma-4gemma-4-12b-itgooglemultimodalopen-sourceai-model

DISCOVERED

1h ago

2026-06-05

PUBLISHED

2h ago

2026-06-05

RELEVANCE

8/ 10

AUTHOR

finguru1980

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

MODEL30m ago

Google releases Gemma 4 QAT checkpoints

Google DeepMind has released official Quantization-Aware Training (QAT) checkpoints for the Gemma 4 model family on Hugging Face, integrating model compression directly into the training process. The release includes unquantized Q4_0 checkpoints, GGUF formats, a mobile-optimized wNa8o8 schema, and compressed tensors for native vLLM inference.

UPDATE32m ago

OpenCode 1.16.2 adds Snowflake Cortex, durable epochs

OpenCode v1.16.2 updates the open-source AI coding assistant with durable context epochs for v2 sessions and multi-server state isolation in the desktop app. The release also integrates Snowflake Cortex and enhances the terminal UI to support direct diff hunk and subagent task navigation.

UPDATE50m ago

Warp Agent adds terminal message queueing

Warp Agent has introduced a message queueing feature for terminal sessions, enabling users to queue, reorder, and auto-queue prompts. Developers can orchestrate a pipeline of commands and prompts for the agent to process sequentially, streamlining debugging and setup workflows.