Gemma 4 26B A4B IT

Gemma 4 26B A4B IT is Google's instruction-tuned mixture-of-experts (MoE) model, built to provide strong text and image understanding while activating only a subset of its parameters for each token.

Capabilities and context limits

The model contains approximately 25.2 billion total parameters but activates roughly 3.8 billion during inference. Google's Gemma 4 model card describes it as an efficient alternative to the dense 31B variant. It accepts text and images and generates text, supporting instruction following, coding, reasoning, and image analysis.

The Gemini API model listing reports 262,144 input tokens and 32,768 output tokens. These limits are useful for long prompts, although actual performance depends on input complexity and provider quotas. The 26B A4B variant does not support native audio input or audio output.

API access and pricing

The model's Gemini API identifier is gemma-4-26b-a4b-it. Google's Gemma API guide explicitly demonstrates text generation with this ID and the generateContent endpoint. EidoStack users can choose it under Google models, provide their own API key, and compare its answers with other models.

Google currently lists Gemma 4 input and output tokens as free on the Gemini API pricing page, with no published paid-tier token pricing. Rate limits and eligibility are determined by Google.

When to use this model

The mixture-of-experts design makes Gemma 4 26B A4B IT a practical candidate for interactive prompt experiments and coding tasks where response speed matters. Compare it with the 31B dense model on representative prompts, inspecting correctness, generated tokens, and latency rather than assuming that the smaller active parameter count determines output quality.

The API-specific token limits above are from Gemini API model-list metadata published in the Google AI Developers Forum.