Models·Google

Gemma 4 26B A4B (free)

google/gemma-4-26b-a4b-it:free

Gemma 4 26B A4B IT is an instruction-tuned Mixture-of-Experts (MoE) model from Google DeepMind. Despite 25.2B total parameters, only 3.8B activate per token during inference — delivering near-31B quality at...

FreeInput, per million tokens
FreeOutput, per million tokens
262KContext window
33KLongest answer

262K tokens of context — about 196,608 words, on a scale that ends at 2M.

What it can do

  • Reasoning
  • Vision
  • Tool use
  • Open weights
  • Free
  • Takes image
  • Takes text
  • Takes video
  • Gives text

Released 3 Apr 2026 · checked 3 Oct 2026.

About it

Gemma 4 26B A4B is Google’s instruction-tuned Mixture-of-Experts model, with open weights. The listed summary says it has 25.2 billion total parameters and activates 3.8 billion per token during inference. It accepts image, text, and video inputs and returns text. Tool use, vision, and reasoning are supported. Its context window is 262,144 tokens, and its maximum output is 32,768 tokens. The listed model ID is google/gemma-4-26b-a4b-it:free, while the Hugging Face ID is google/gemma-4-26B-A4B-it. Google released it on 2026-04-03. It is marked free, with no per-token input or output prices listed. Teams evaluating an open-weight model can consider its supported inputs, context, and output bounds alongside its instruction-tuned and MoE characteristics.

Verdict

Gemma 4 26B A4B is an open-weight model with multimodal inputs, tool use, and reasoning support. Its 262,144-token context and 32,768-token output limit define useful boundaries for evaluating it.

  • ["Open weights are available.", "Accepts image, text, and video.", "Supports tool use, vision, and reasoning.", "262,144-token context window.", "Marked free."]
  • ["Maximum output is 32,768 tokens.", "No per-token input or output prices are listed."]

On these boards

More from Google