Overview
Gemma 4 E4B IT Assistant is an open-weight Multi-Token Prediction (MTP) drafter for the instruction-tuned Gemma 4 E4B target model. Its release repository is google/gemma-4-E4B-it-assistant; this is a checkpoint identifier, not a verified API model ID. The model card lists Apache 2.0 licensing. (huggingface.co)
Capabilities and uses
- The drafter proposes multiple candidate tokens for the target model to verify in parallel, using the target's activations and shared context information. It is intended to accelerate generation, not to replace the target model as an independent assistant. (ai.google.dev)
- Google reports up to 3× faster generation for the Gemma 4 MTP family in tested configurations, without output-quality degradation. Actual speed depends on the model, hardware and inference setup; this is not a guaranteed E4B-specific result. (blog.google)
- Paired with E4B, it can support lower-latency on-device applications, coding assistants and agent workflows. The target E4B model supports text, image and audio input and text output, with a 128K-token context window; these are target-model capabilities, not independent drafter capabilities. (ai.google.dev)
Limitations
The checkpoint requires its matching target model and a compatible speculative-decoding implementation. Faster inference does not remove the target model's risks, including incorrect or outdated answers, bias and difficulty with complex or ambiguous tasks. (ai.google.dev)
Release date
Google's announcement is dated May 5, 2026. Its release-history page separately lists Gemma 4 MTP, including E4B, under April 16, 2026; the official sources therefore report different dates. (blog.google)
Pricing
As of November 16, 2026, a separate official hosted API price for this assistant checkpoint could not be verified: input and output token rates and currency are unverified. Google's Gemini API documentation lists other Gemma 4 variants, not this assistant checkpoint, so their pricing should not be attributed to it. Self-hosting can incur infrastructure costs. (ai.google.dev)