Explore additional models
Frequently
asked questions
Gemma 4 26B A4B is good for multilingual chat, summarization, instruction following, and assistant workloads that need fast responses at high concurrency. A mixture-of-experts model with 25.2B total parameters and 3.8B active per token, it accepts text and image input for document and visual understanding, with a 262,144-token context. It is available for Self-Serve Deployments on dedicated GPUs in Crusoe Intelligence Foundry.
Gemma 4 26B A4B is a mixture-of-experts model that activates 3.8B of its 25.2B parameters per token. Gemma 4 31B IT is a dense model that uses all of its parameters on every token. The 26B A4B needs less compute per token, so it generates faster and serves more requests per GPU. The 31B scores higher on quality: at launch it placed third among open models on LMArena’s text leaderboard, and the 26B A4B placed sixth. On Crusoe, the 26B A4B is available for Self-Serve Deployments, and the 31B IT is available for Serverless Inference, Self-Serve Deployments, and Serverless Fine-Tuning. You can also request a Tailored Deployment for bespoke inference optimization and a dedicated endpoint.
Gemma 4 26B A4B is available for high-performance inference via Self-Serve Deployments, billed per GPU hour with no per-token charges. You get a dedicated endpoint with an OpenAI-compatible API in Crusoe Intelligence Foundry. Serverless Inference and Serverless Fine-Tuning are not currently available for this model. Current rates are shown in the pricing module above.
A4B stands for about 4 billion active parameters. Gemma 4 26B A4B spreads 25.2B parameters across 128 experts, and for each token a router activates 8 experts plus 1 shared expert, about 3.8B parameters in total. The full parameter count sets how much GPU memory the model needs. The active count sets how much compute each token takes, which is why it runs faster than a dense model of the same size.
Gemma 4 26B A4B is not currently available for Crusoe Serverless Fine-Tuning. Its Apache 2.0 license permits fine-tuning and commercial use, so you can fine-tune it on your own infrastructure, including Crusoe Cloud GPU instances. For managed LoRA fine-tuning, Gemma 4 31B IT is available for Serverless Fine-Tuning and deploys in one click using Self-Serve Deployments for high-performance inference.
MemoryAlloy technology is a proprietary memory fabric in Crusoe’s inference engine that shares the KV cache across nodes. Cache-aware routing sends each request to the node most likely to have its context ready. For Gemma 4 26B A4B, repeated system prompts, shared documents, and multi-turn conversations skip redundant prefill, so each deployment handles more concurrent requests at consistent latency. Self-Serve Deployments bill per GPU hour, so more requests per GPU lower your cost per request.


%204.png)
