Positioning
Gemma 4 E2B IT Assistant is an open-weight, four-layer multi-token prediction (MTP) drafter for the instruction-tuned Gemma 4 E2B target model. Its repository identifier is google/gemma-4-E2B-it-assistant; it is not a standalone general-purpose chat model. The model card lists an Apache 2.0 license. (ai.google.dev)
Capabilities and uses
- It proposes candidate tokens that the target model verifies through speculative decoding. It shares the target model’s input embeddings and KV cache and uses its final-layer activations. The E2B drafter also uses token clustering to reduce prediction computation. (ai.google.dev)
- It is intended to reduce generation latency in applications such as local and on-device inference. Google reports up to a 3× speedup for the Gemma 4 MTP family in its tests, without degradation in output quality; this is not a guaranteed E2B result. (blog.google)
- Google’s example pairs this checkpoint with
google/gemma-4-E2B-itusing a compatible Transformers speculative-decoding workflow. The target model handles the application’s inputs and final output; the drafter’s role is to accelerate generation. (ai.google.dev)
Limitations
The drafter depends on a compatible target model and runtime, so it should not be treated as an independent multimodal assistant. Actual speedup depends on hardware, runtime and workload. The target model’s limitations, including possible errors and biases, still apply; speculative decoding is an inference optimization rather than a guarantee of factual correctness. (ai.google.dev)
Dates
Google’s announcement is dated May 5, 2026. Its Gemma release log separately lists the E2B MTP family release under April 16, 2026; the two official dates should not be assumed to describe the same milestone. (blog.google)
Pricing
A separate official API input-token or output-token price for this specific assistant checkpoint could not be verified. Google’s Gemini API pricing page lists a free tier for Gemma 4 generally, but does not establish that this drafter checkpoint is separately available at that price. Local deployment can incur infrastructure costs. Pricing checked November 16, 2026. (ai.google.dev)