Google Gemini blitz adds Gemma 4, Lyria 3 and new AI tiers

Google’s latest Gemini blitz shows just how fast the frontier platform layer is fragmenting and recombining. In four days it shipped a new open‑weight family (Gemma 4), a cost‑optimized video model (Veo 3.1 Lite), a higher‑fidelity music generator (Lyria 3), a multimodal embedding model and new inference pricing tiers—not to mention another round of deprecations for older Gemini endpoints. The message is clear: Google wants to own every modality and price point in the stack, from local Gemma deployments up through full Gemini APIs.

For the race to AGI, this cadence matters in two ways. First, it keeps pressure on OpenAI, Anthropic and Chinese labs to match not just headline model quality but also packaging—embeddings that handle all media types, unified billing, and flexible latency/cost trade‑offs. Second, the relentless deprecation treadmill raises the operational bar for anyone building on frontier APIs: if preview models live for only 4–8 weeks, serious teams must treat model versioning and migration as a core competency, not an afterthought.

Strategically, Gemma 4 is the quiet but important piece. By putting a high‑end, sparsely activated open model under Apache‑style terms, Google blurs the line between closed and open ecosystems. Combined with Lyria and Veo, it’s building an end‑to‑end creative and agentic stack that can run partly on‑prem and partly in Google Cloud, giving enterprises a credible alternative to OpenAI‑centric architectures.