Google · Google (Gemini / Gemma / Imagen / Veo)
Gemma 4 (27B)
ActiveOpen-weight family (Hugging Face) for self-hosting/on-device; flagship 27B MoE (~14B active) fits ~14GB VRAM, 256K context.
About this version
Gemma 4 is the latest open‑weight model in the Gemini family, released as a 27‑billion‑parameter mixture‑of‑experts architecture that typically activates around 14 billion parameters and can run on approximately 14 GB of VRAM. It supports an extended context window of up to 256 000 tokens and is intended for self‑hosted or on‑device deployment via the Hugging Face ecosystem, continuing the family’s focus on accessible, high‑capacity language models.
- Context window
- 256K ctx
- Tier
- Open Weight
Release & change timeline
No dated events logged for this version yet.
Patches, price changes, benchmarks, and deprecations for Gemma 4 (27B) will appear here as they’re recorded. Follow the broader Catch Me Up digest in the meantime.
Report an issue with Google (Gemini / Gemma / Imagen / Veo)
Posts to your status feed
Pick the closest match below, edit the body, and post. Your report carries the #gemini tag automatically so it surfaces here + in the trending-tags rail. Reporting also follows Google (Gemini / Gemma / Imagen / Veo) so you’ll get status updates.
Part of the Google (Gemini / Gemma / Imagen / Veo) version timeline. Catch Me Up — everything that changed across AI & smart home ↗