Fireworks is now a first-party inference provider in Microsoft Foundry.
Billed through your existing Azure account.
TRUSTED BY LEADING AI TEAMS
“By running Fireworks on Azure Foundry, UiPath powers both Autopilot and Delegate with open models that are significantly faster and more cost-efficient for Computer Use, all while matching the quality of Claude’s Sonnet 4.6. It’s a step-change in how we deliver AI at scale to our customers.”
Microsoft CVP Yina Arenas recently sat down with Fireworks CEO Lin Qiao to discuss Fireworks on Foundry.
Azure-native procurement, compliance, security, and dedicated capacity, now combined with Fireworks inference. All inside the environment you already operate.
First-party in Microsoft Foundry. Identity through Entra ID, billing through Azure, governance through your tenant.
20+ frontier open models: DeepSeek, Kimi, GLM, gpt-oss, MiniMax. All accessible through one API.
PTU-backed dedicated capacity, OpenAI-compatible API, and the Fireworks inference engine. Production-grade out of the box.
Usage runs through your existing Azure account and counts toward MACC. PTU is also ACD-eligible. Security policies, IAM, and audit inherited from your tenant.
The Fireworks inference engine and FireOptimizer compilation are tuned per model architecture for high throughput and low latency in production.
Up to 1M TPM on serverless PayGo, with dedicated PTU capacity for production workloads. All inside your Azure subscription.
Import and deploy your own model weights on Foundry using the Fireworks inference runtime.
Multi-tenant serverless for prototyping and bursty workloads, or single-tenant dedicated capacity for production. Both billed through Azure.
Serverless · Multi-tenant
Per-token pricing with no commitment. The fastest way to get started. Perfect for startups, prototyping, and bursty workloads.
PRICING: Per token · MACC-eligible
REGIONS: 6 US Data Zone regions
THROUGHPUT: Up to 1M TPM
SETUP:
Single-tenant · Dedicated capacity
Dedicated GPU capacity reserved for your workload. Consistent throughput, stable endpoints, broader region availability. Best for production applications with predictable traffic.
PRICING: Per PTU-hr · ACD + MACC
REGIONS: Global (subject to capacity)
CAPACITY: Dedicated · min 2 replicas
SETUP:
DeepSeek, Kimi, GLM, MiniMax, gpt-oss, etc. are each served via the OpenAI-compatible Chat Completions and Responses APIs. New models added on a rolling basis.
Fireworks provides a production-grade inference engine built specifically for modern open-weight models, now available natively within Microsoft Foundry.
Run your proprietary or fine-tuned open-weight models within the Foundry ecosystem using the Fireworks inference runtime.
Usage is billed through your Azure account and is MACC-eligible. PTU is also ACD-eligible. Identity via Entra ID, governance from your tenant.
Triple security certification. Zero data retention by default. Prompts exist only in volatile memory for the duration of each request.
Fireworks runs as a first-party provider inside Microsoft Foundry under Microsoft’s terms, not a third-party vendor your procurement needs to onboard.
For full setup steps and operational detail,
see the Microsoft Learn enable guide.
First, make sure you’re on the New Foundry experience (ai.azure.com/nextgen). Then in the Azure Portal go to Subscriptions → Preview features → search Fireworks.Enable.Deploy → Register. Propagates in up to 30 minutes. Full setup guide on Microsoft Learn →
Default PayGo capacity is up to 1M TPM per request. Higher quotas are available. Request at aka.ms/fireworks-quota. Quota is granted across all Fireworks models on your subscription, not per-model.
Not today. This is a current capacity limitation. PayGo is available in US Data Zone regions; PTU is global subject to capacity. EU coverage is on the roadmap. Reach out to azure@fireworks.ai if EU residency is a hard requirement and we’ll work through it with you.
Azure credits can be applied to PTU deployments. PayGo usage is billed at standard token pricing and counts toward MACC.
OpenAI-compatible Chat Completions and Responses APIs. If your code already uses the OpenAI SDK, you only need to change the base URL and key.
PayGo: 6 US Data Zone regions (East US, East US 2, Central US, North Central US, West US, West US 3). PTU is available globally subject to capacity. FedRAMP and PCI are not currently certified.
New models go through a Microsoft onboarding pipeline, typically 5–10 days after we release them on the Fireworks platform. For day-of-release access, use fireworks.ai directly.
Access Fireworks directly from Microsoft Foundry. Usage is billed through your Azure account and is MACC-eligible.
PTU provisioning or BYOM questions? Contact azure@fireworks.ai
