Official DeepMind & Google Blog RSS + GitHub Releases + Hugging Face.
71 updates · page 1 of 5
Google DeepMind describes using Google Antigravity for Science to speed the search for novel antimicrobial peptides.
Why it matters: Scientists hunting antimicrobial peptides can apply Antigravity for Science to that search.
Google GenAI SDK (Python) v2.31.0 adds gemini-4-argon to the Model enum and fixes continuation state, union-type function declarations, and grounding metadata merging.
Why it matters: Python developers on this SDK can select gemini-4-argon and get the continuation and parser fixes in v2.31.0.
Google DeepMind and GovTech Singapore are collaborating on a fine-tuned model that uses lightweight on-device Gemma models to intercept conversational scams in private chats with zero cloud data egress.
Why it matters: Private chats can be checked for conversational scams on device, with no cloud data egress.
Gemini CLI v0.64.0 includes a2a-server settings migration from V1 to V2 and bridges ACP prompt usage so usage_update notifications can be emitted.
Why it matters: People on Gemini CLI can upgrade to v0.64.0 for that settings migration and the usage updates in this release.
Google DeepMind published a post on developing Gemini 4 Argon responsibly and safely.
Why it matters: Teams looking at Gemini 4 Argon can read how DeepMind is approaching safety for this model.
Google DeepMind published Gemini 4 Argon for Robotics.
Why it matters: Robotics teams can read how Gemini 4 Argon is being applied to robots.
Google is giving developers, enterprises, scientists, and consumers access to Gemini 4 Argon, its latest frontier model.
Why it matters: Those groups can now use Gemini 4 Argon on the hardest problems the post describes.
Gemini 4 Argon is now available for developers to build complex software and cybersecurity workflows.
Why it matters: Developers can start from the Gemini API Developer Guide and the Google Antigravity and Android Studio material in the post.
Google GenAI SDK (Python) v2.30.1 fixes concatenation of output_text across multi-step model responses in a turn.
Why it matters: Python callers on v2.30.1 get output_text joined when a turn has more than one model response.
Google GenAI SDK (Python) v2.30.0 adds streaming continuation usage metadata, a continuation token on interaction status updates, and a CONTINUATION_REQUIRED interaction status.
Why it matters: Python developers can track continuation usage and handle CONTINUATION_REQUIRED interactions in v2.30.0.
Google GenAI SDK (Python) v2.29.0 can set an environment network allowlist to disabled, and it logs a direct AFC warning only when functions are present.
Why it matters: Python callers can disable the network allowlist and avoid that warning when no functions are configured.
Google DeepMind introduced Playground, an experimental gaming platform for creating, playing, and sharing custom games.
Why it matters: Players can make a custom game in Playground and share it for others to play.
Gemini CLI v0.63.0 shows retry progress during connection recovery, separates a missing MCP enablement config from malformed JSON, and restores paused stdin after capability detection.
Why it matters: CLI users can watch recovery and avoid treating a missing MCP setting as broken JSON.
Google DeepMind introduced EmbeddingGemma 2, an open multimodal embedding model optimized for privacy-first use cases.
Why it matters: Developers who need on-device or private embeddings can try the open EmbeddingGemma 2 model.
Google GenAI SDK (Python) v2.28.0 adds lyria-3.5, gemini-omni-1.1-flash, and gemini-omni-flash-preview to the Interactions model enum, and supports continuation_token in Interactions.
Why it matters: Developers can call those models through Interactions and continue a run with continuation_token.