Google: Gemma 4 26B A4B (free)

google/gemma-4-26b-a4b-it:free · GoogleDeclared and measured agree

Tokenizer lineage (measured): Gemma
Evidence confidence: High
Declared tokenizer tag: Gemma
Entered catalogue: 3 April 2026
Last tested: Not yet measured
Measurement: Variant: inherits its base model’s measurement

Evidence

✓ inherited from google/gemma-4-26b-a4b-it: a serving variant shares its base model's tokenizer

This entry is a serving variant of Google: Gemma 4 26B A4B . A variant shares its base model's tokenizer by construction, so the measurement lives on the base record and this page inherits it.

Identity statement

Google: Gemma 4 26B A4B (free) is Google's model. Its measured tokenizer sits with the Gemma group, which is evidence that it builds on the Gemma tokenizer. Tokenizer reuse is normal engineering, open vocabularies travel between labs, and this says nothing against Google's authorship of the model itself.

Evidence stack

Signals of different kinds, weighed together

Signal	Result	Reference
Tokenizer signature	Not measured	–
API surface	No other entry shares this exact contract	api_32dbb075
Context and output	262,144 context · 32,768 max output	declared
Reasoning contract	Optional	declared
Serving providers observed	None recorded yet	–

Declared record

What the catalogue claims about this model

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...

Modalities: text+image+video->text
Declared tokenizer: Gemma
Prompt price: $0 / M tokens
Completion price: $0 / M tokens

Supported parameters

include_reasoningmax_tokensreasoningresponse_formatseedtemperaturetool_choicetoolstop_p

defaults: {"temperature":1,"top_p":0.95,"top_k":64}

Catalogue entry Weights on Hugging Face

History

Every observation, kept as taken

3 April 2026Enters the OpenRouter catalogue declared as Gemma.

Cite this page as the evidence record for Google: Gemma 4 26B A4B (free): the URL is stable, measurements are dated, and revisions append to the history above. Method: tokenizer fingerprinting, v1, 50 strings. The raw data behind every figure is on the data page.