Crusoe Managed Inference

Inference
that scales

Serve models with ultra-low latency and high throughput. Now with flexible deployment options that fit your workload.

Trusted by:

Crusoe's inference engine is powered by MemoryAlloy™ technology, a unique cluster-native memory fabric that persists across nodes and enables intelligent routing to eliminate redundant prefill computation.

Abstract illustration composed of layered curved lines forming an organic shape

Flexible deployment options

Serverless Inference

Get started fast with usage-based consumption of top open-source models via a fully managed inference API in Crusoe Intelligence Foundry.

Best for

• Early-stage workloads

• Low volume traffic

• Rapid experimentation

Self-Serve Deployments

Run top models in Crusoe Intelligence Foundry with inference optimized for throughput or responsiveness. Deploy models customized with Serverless Fine-Tuning in one click.

Best for

• Production-scale workloads

• Low-latency applications

• Predictable costs

Tailored Deployments

Work with our team for the highest level of optimization and a dedicated, benchmarked endpoint. You bring the model and define the requirements, we handle the rest.

Best for

• Proprietary models

• Bespoke inference optimization

• SLA-backed performance

Request benchmarking

Model hub

Explore a curated selection of pre-configured top open models from leading AI labs, ranging from lightweight and ultra-fast to large with frontier reasoning power.

This is some text inside of a div block.

models

Model

Parameters

Context

AVAILABLE FOR

Qwen3 235B A22B Instruct

Parameters

235000000000

Context

262144

AVAILABLE FOR

Self-Serve Deployments

Qwen3.5 9B

Parameters

9000000000

Context

262144

AVAILABLE FOR

Self-Serve Deployments

Qwen3.6 27B

Parameters

27000000000

Context

262144

AVAILABLE FOR

Self-Serve Deployments

Qwen3.5 2B

Parameters

2000000000

Context

262144

AVAILABLE FOR

Self-Serve Deployments

Qwen3.5 35B A3B

Parameters

35000000000

Context

262144

AVAILABLE FOR

Self-Serve Deployments

GLM-5.2

Parameters

753000000000

Context

262144

AVAILABLE FOR

Self-Serve Deployments

Gemma 4 26B A4B

Parameters

25200000000

Context

262144

AVAILABLE FOR

Self-Serve Deployments

Qwen3 8B

Parameters

8200000000

Context

40960

AVAILABLE FOR

Self-Serve Deployments

Qwen3.6 35B A3B

Parameters

35000000000

Context

262144

AVAILABLE FOR

Self-Serve Deployments

Llama 3.3 70B Instruct

Parameters

70000000000

Context

131072

AVAILABLE FOR

Self-Serve Deployments

GPT-OSS 20B

Parameters

21000000000

Context

131072

AVAILABLE FOR

Self-Serve Deployments

Nemotron 3.5 Super VL 120B A12B

Parameters

120000000000

Context

262144

AVAILABLE FOR

Serverless Inference

GLM-5.3 Flash

Parameters

321000000000

Context

1048576

AVAILABLE FOR

Serverless Inference, Self-Serve Deployments

GLM-5.3

Parameters

753000000000

Context

1048576

AVAILABLE FOR

Serverless Inference, Self-Serve Deployments

DeepSeek V4.1 Flash

Parameters

552000000000

Context

1048576

AVAILABLE FOR

Serverless Inference

DeepSeek V4 Flash 0731

Parameters

284000000000

Context

1048576

AVAILABLE FOR

Serverless Inference, Self-Serve Deployments

DeepSeek V4 Pro

Parameters

1600000000000

Context

1048576

AVAILABLE FOR

Serverless Inference

Gemma 4 31B

Parameters

30700000000

Context

262144

AVAILABLE FOR

Serverless Inference, Self-Serve Deployments

GPT-OSS 120B

Parameters

117000000000

Context

131072

AVAILABLE FOR

Serverless Inference, Self-Serve Deployments

Llama 3.1 8B Instruct

Parameters

8000000000

Context

131072

AVAILABLE FOR

Self-Serve Deployments

Nemotron 3.5 Lightning 30B A3B

Parameters

30000000000

Context

262144

AVAILABLE FOR

Serverless Inference, Self-Serve Deployments

Nemotron 3 Nano 30B

Parameters

30000000000

Context

262144

AVAILABLE FOR

Serverless Inference

Nemotron 3 Nano Omni

Parameters

31000000000

Context

262144

AVAILABLE FOR

Serverless Inference

Nemotron 3 Super 120B

Parameters

120000000000

Context

262144

AVAILABLE FOR

Serverless Inference

Nemotron 3 VoiceChat

Parameters

11000000000

AVAILABLE FOR

Serverless Inference

Qwen3.8 27B

Parameters

27000000000

Context

262144

AVAILABLE FOR

Serverless Inference, Self-Serve Deployments

Unmatched performance, flexibility, and scalability

1

Breakthrough speed

Powered by MemoryAlloy technology with a cluster-wide KV cache, Crusoe's inference engine delivers up to 9.9x faster time-to-first-token* and 5x higher throughput using speculative decoding and dynamic batching.

2

Ultimate flexibility

Choose the right inference deployment option for your workload. Pay as you go with Serverless Inference APIs, use Self-Serve Deployments to further optimize for throughput or responsiveness, or work with us via Tailored Deployments to customize for your unique needs.

3

Seamless scaling

Go from experiment to production without the operational burden. We handle the infrastructure to deliver reliable performance even under peak load. Meet changing user demands with elastic scaling that flexes from single-user testing to sustained, high-volume production deployments.

*

Benchmarked against vLLM for Llama-3.3-70B model. Read our blog to learn more details.

In voice, inference latency is the product; users hear every millisecond you didn't optimize. Crusoe tuned our stack against real production traffic, and the payoff is conversation that flows at human speed.

Eric Chen

VP of AI

Our biggest fear was that cutting costs would cut the product quality. With Crusoe Managed Inference, performance and latency improved, and we're paying an order of magnitude less.

Maor Shlomo

Founder

Crusoe Intelligence Foundry, designed for AI developers

Discover new models, accelerate fine-tuning, and quickly deploy with flexible inference options.

API keys for fastest path to production

Experiment with top open-source models rapidly. Generate API keys, monitor performance metrics and enable provisioned throughput for production-scale deployments.

Managed endpoints for rapid deployment

Leverage fully managed endpoints powered by our inference engine tuned specifically to each model for optimized performance.

Unified interface for cross-team collaboration

Users working across teams can easily switch between the Crusoe Intelligence Foundry for inference tasks and the Crusoe Cloud Console for infrastructure-as-a-service (IaaS) resources within a single, integrated environment.

Crusoe inference engine vs vLLM

0

2

4

6

8

10

TTFT

Throughput

9.9x

5.0x

x Improvement vs. vLLM

Llama-3.3-70B model, 4-node deployment

$Screenshot of a coding interface showing a prompt requesting a Python 3.11+ fractal viewer app, followed by a partial Python script with imports and comments.$

Frequently
asked questions

Are you ready to build something amazing?