← All open-weight models

Google DeepMind · Vision & multimodal

Gemma 4 26B A4B Instruct

Best open-weight throughput-per-dollar in the family — MoE activating only 4B params per token, so it serves near-4B-dense speed at 26B-class quality; free-tier only on the hosted Gemini API, weights on Kaggle/Hugging Face with QAT quantized builds.

Open weights · Community or custom license · Data checked September 6, 2026 · Not a hands-on evaluation

Model ID gemma-4-26b-a4b-it

Parameters
26B total · 4B active26B total / 4B active MoE
Architecture
Mixture of experts
Context window
262,144 tokensMaximum input
Max output
Not recordedPer response
Input price
$0.00 per 1M tokens
Output price
$0.00 per 1M tokens
Cached input
Not published
Blended price
$0.00 per 1M tokens3:1 input to output
Knowledge cutoff
Not recorded
Released
2026
Status
ga
Size band
15B to 40B

License and openness

Gemma 4 26B A4B Instruct is released under the Gemma 4 license (Gemma Terms of Use). This is a community or custom license with its own conditions, which can include acceptable-use rules, user thresholds or naming requirements. Read it before commercial use. License terms can change between versions; confirm the text published with the weights you download.

Capabilities

Input modalitiestext, image
Output modalitiestext

What it is for

At a three-to-one input-to-output ratio, Gemma 4 26B A4B Instruct costs $0.00 per million tokens blended. A workload of one million input and 330,000 output tokens per day would run about $0.00 per month at list price, before caching or batch discounts.

Cost at list price

Input$0.00 per 1M tokens
Output$0.00 per 1M tokens
Cached inputNot published
Blended, 3:1$0.00 per 1M tokens

A recorded $0.00 is preserved from the reference as recorded. It does not promise free access; check the provider’s current terms.

Estimate a workload across all models ↗

Sources

Sources checked September 6, 2026. Specifications and prices change without notice; confirm against the provider before you commit.

Other open-weight models from Google DeepMind

Browse all open-weight models ↗