Gemma 4: April 2026 Open Models and the Separate June 12B Multimodal Release. This reference distinguishes the original release from hosted endpoints, downloadable weights and subsequent version changes. It is written for people deciding whether an exact model or checkpoint is suitable for a real application, not for leaderboard marketing.
What the sources actually establish
Google introduced Gemma 4 on April 2, 2026 under Apache 2.0. That family launch is not the same as its June 3 announcement of Gemma 4 12B, a later unified encoder-free multimodal variant with native audio input. An accurate timeline must therefore distinguish the initial family, the 12B checkpoint and any subsequently published quantization-aware or draft-model versions.
A useful application and evaluation scenario
For a laptop document assistant, compare the 12B variant with a smaller Gemma checkpoint using the identical sample of licensed documents, short audio clips and image-heavy pages where applicable. The 12B announcement described operation on devices with 16GB of RAM or unified memory; that is a vendor configuration claim, not a guarantee that an arbitrary 16GB machine can run the model alongside a full desktop workload.
Limits, risks and uncertainties
Multimodal inputs depend on the selected exact variant; do not label every Gemma 4 checkpoint as audio-capable. Apache 2.0 is stated for this model family, but packaging, third-party datasets, redistribution and hosted endpoint conditions still merit project-level review. Responses can misread tables or speech; local control of weights does not eliminate error.
Alternatives and the decision rule
Choose a smaller Gemma 4 model for constrained-device latency; test 12B when speech-plus-vision is core to the user journey. A hosted API may suit bursty demand better than maintaining GPU machines, while local inference offers control over input handling subject to device security. For image or voice production workflows, link to AI03–05 instead of treating text understanding as media generation.
A reproducible test before adoption
Prepare a small held-out set from data the organization is authorized to process. Include simple, ambiguous and deliberately unanswerable requests; specify the expected answer or evidence and a human-reviewed scoring rule. Record model identifier, endpoint and region, tokenizer or SDK version, prompt, tool settings, output budget and whether external retrieval was allowed. Do not present a provider-supplied benchmark as an independently measured result. Separate model mistakes from retrieval, OCR and application validation failures.
Operational checklist and dated evidence
A production transition must test 429 responses, timeouts, refusals, invalid structured output and model-unavailable errors. Keep a reversible rollout or safe fallback instead of silently substituting a model behind an old identifier. Measure latency and usage on the actual workload before planning budgets; join dated token quantities to the AI 10 price ledger. Verify enterprise residency, region, support entitlement and data retention with the actual contract, not a public release story.
Sources & Further Reading
- blog.google — documentation/reference (supplier-provided reference; not independently checked in this pass)
- blog.google — documentation/reference (supplier-provided reference; not independently checked in this pass)
Source documentation establishes vendor-published facts, not independently measured model quality, tenant access, contractual rights or current prices. Use a dated pricing ledger and an account-specific entitlement check for deployment decisions.