Crusoe Updates: October 2026

Track model and hardware releases, API changes, and feature launches.

Managed AI

October 21, 2026000120Models
DeepSeek V4.1 Flash updated to the 1017 checkpoint
DeepSeek V4.1 Flash has been refreshed with a new checkpoint. The deepseek-ai/DeepSeek-V4.1-Flash model now serves the updated weights in place. See Available models.
October 8, 2026000119Models
New model available: DeepSeek V4.1 Flash
Intelligence Foundry now supports DeepSeek's V4.1 Flash model, a 763B-parameter instruct model with a 1M-token context window. See Available models.
October 2, 2026000116ModelsBreaking
Serverless Inference model deprecation
Moonshot Kimi K2.6 was deprecated for Serverless Inference on October 2, 2026. See Migrate from a deprecated model.

October 21, 2026000121Major release
Multi-LoRA for Self-Serve Deployments
You can now deploy a base model with multiple adapters to test and compare checkpoints side-by-side on the same GPUs. Deploy a base model once and hot-load several LoRA adapters at runtime, with each adapter exposed as its own model ID on the same OpenAI-compatible endpoint. See Self-Serve Deployments.

October 7, 2026000118Service update
Crusoe Cloud Interconnect
You can now obtain a dedicated connection to a Crusoe Cloud VPC to send traffic from a point of presence to your VPC's region. This ensures that your traffic never crosses the public internet. See Interconnect locations.

October 5, 2026000117Security
GCP Cloud KMS support for customer-managed encryption keys
Customer-managed encryption keys (CMEK) now support GCP Cloud KMS in addition to AWS KMS. Register a Cloud KMS key through workload identity federation to encrypt and decrypt your data with a key you own in your GCP project. See Customer-Managed Keys.