| CARVIEW |
Your compute sanctuary
Your compute sanctuary
Rest easy knowing you can get the compute you need for any AI workload
.
Customers and partners
Customers and partners
Trusted by AI-native startups, research labs, public companies, and academic institutions.
Developers of EVO, a genomic foundation model
“Mithril’s omnicloud platform has accelerated science at Arc.
Our machine learning work brings demanding performance infrastructure needs, and Mithril delivers. With Mithril, we can guarantee that our researchers have exactly the compute they need, when they need it, without procurement friction.”
Patrick Hsu
Co-founder and CEO, Arc Institute
Omnicloud AI compute
Mithril aggregates and orchestrates multi-cloud GPUs, CPUs, and storage so you don't have to. Access the infrastructure you need through a single platform with transparent pricing.
1
Burst the compute you need
Prevent overprovisioning by spinning up GPUs via extendable short-term reservations or spot bids.
2
Pay the market rate
Get the best compute prices without talking to 10 sales teams. Mithril sets compute prices across all capacity algorithmically, based on supply and demand.
3
Run batch workloads
Use the batch SDK and batch inference API to run workloads asynchronously on spot capacity at the best price.
Compute instances
Compute
instances
Batch inference
Batch
inference
Cloud
Cloud
Cloud
Datacenter
Datacenter
Datacenter
Datacenter
Datacenter
Datacenter
Compute instances
Compute instances
Compute the way you need it
Compute the way you need it
Spin up GPUs through a simple, transparent platform
Training
Fine-tuning
Inference
Batch jobs
Training
Fine-tuning
Inference
Batch jobs
Spin up GPUs for hours, days, or months
Spin up GPUs for hours, days, or months
Reserve what you need in minutes
Instantly access GPU VMs with extendable short-term reservations.
Don't overpay
See pricing for all compute in real-time. Get the best prices without negotiating.
Scale to thousands of GPUs
Optimize your economics for large-scale clusters without shopping around every cloud.
Reserve what you need in minutes
Instantly access GPU VMs with extendable short-term reservations.
Don't overpay
See pricing for all compute in real-time. Get the best prices without negotiating.
Scale to thousands of GPUs
Optimize your economics for large-scale clusters without shopping around every cloud.
Burst spot instances for async workloads
Burst spot instances for async workloads
Burst when the price is right
Limit spend by creating spot bids that provision GPUs when they're available at your desired price.
Run batch jobs for less
Queue up async jobs to run on spot compute programmatically.
Pause and resume
Pause your bids without losing your environment whenever you don't have workloads to run.
Burst when the price is right
Limit spend by creating spot bids that provision GPUs when they're available at your desired price.
Run batch jobs for less
Queue up async jobs to run on spot compute programmatically.
Pause and resume
Pause your bids without losing your environment whenever you don't have workloads to run.
Purpose-built infrastructure for machine learning workloads
Purpose-built infrastructure for machine learning workloads
NVIDIA GPUs with InfiniBand
Run distributed workloads on NVIDIA A100s, H100s, and H200s with pre-configured interconnect.
High-performance storage
Maximize workload performance with co-located WEKA block and fileshare storage.
API and batch job SDK
Provision compute via API and launch ML batch jobs on spot instances with the Flow SDK.
Streamlined access and security controls
Project-level permissions, global SSH keys for admin monitoring, and SSO for secure access—all backed by SOC 2 compliant infrastructure.
Batch inference
Batch inference
Trillions of tokens, effortlessly
Trillions of tokens, effortlessly
Asynchronously process millions of inference requests with a simple API
Content generation
Data classification
Embeddings
Prompt testing
Inference, orchestrated for efficiency
Inference, orchestrated for efficiency
More tokens for less
Get the best economics for processing massive multimodal datasets.
Start in minutes
Use the OpenAI SDK to easily transfer existing workloads or create new batch jobs.
Any use case
Experiment with text models in minutes. Deploy any open or custom model you need.
Predictable delivery
Simple, token-based pricing for 24-hour job completion. If your job takes longer, it's free.
Models
Models
Text
gpt-oss-120b
MoE · 117B parameters
Text
gpt-oss-20b
MoE · 21B parameters
Text
Llama 4 Maverick 17Bx128E Instruct FP8
MoE · 400B parameters
Text
Llama 4 Scout 17Bx16E Instruct
MoE · 109B parameters
Text
Llama 3.1 8B Instruct
Dense model · 8B parameters
Text
Qwen 2.5 72B Instruct
Dense model · 72B parameters
Custom
[Your model]
Hosted open and custom models