We're excited to expand our offerings for Fireworks AI on Microsoft Foundry with the availability of DeepSeek V4 Pro and Kimi K2.6, as well as expanded US Data Zone PTUs.
We launched the public preview of Fireworks AI on Microsoft Foundry two months ago, bringing best-in-class open model inference directly into Azure. The traction has been real and high. Fireworks AI's inference layer is purpose-built for low-latency, high-throughput production workloads, natively within Azure. If you're building on Azure and want the best open models without standing up your own inference infrastructure, this is built for you.
What’s Driving Adoption
Since launch, Fireworks AI on Foundry is seeing strong customer momentum, with customer count growing week-over-week and deepening enterprise penetration globally. Teams across industries have moved quickly from experimentation to production.
The pattern is consistent: teams want frontier open model performance, without rearchitecting around a new provider, managing compliance separately, or accepting latency trade-offs. Fireworks AI on Foundry is directly addressing it by combining Fireworks’ industry-leading inference engine with the Azure enterprise control plane they already trust.
Watch our leaders share what's driving this momentum — and where we're headed next
PTU Now Available in the US Data Zone
We’re expanding provisioned throughput unit (PTU) support to the US Data Zone, giving US-based customers a path to predictable, steady-state performance with their data staying within US Azure regions. This is a significant step for regulated industries and organizations with data residency requirements who want the consistent SLAs of PTUs alongside the data sovereignty commitments of Azure’s US Data Zone.
PTUs are the right choice when you need consistent token throughput under sustained load. Serverless works great for experimentation; PTUs are for when you're in production and variability isn't acceptable. Enterprise subscriptions already include PTU quota. Need more capacity? Request it via this form.
New Models: Kimi K2.6 and DeepSeek V4 Pro
Two of the most capable new open models are now available through Fireworks AI on Foundry:
Kimi K2.6 — Moonshot AI's latest reasoning model. Strong on complex, multi-step tasks, structured analysis, research assistants, agent workflows that need to plan across long context. Serverless pricing: $0.95/1M input · $0.16/1M cached · $4.00/1M output.
DeepSeek V4 Pro — DeepSeek's flagship coding and reasoning model. If your team is generating code at production scale, this is worth benchmarking. Serverless pricing: $1.75/1M input · $0.15/1M cached · $3.48/1M output.
Both are available per-token serverless and via PTU through the Foundry model catalog with a single Azure endpoint and same enterprise controls. PTU US Data Zone support means you can run either model with predictable performance and data residency alignment out of the box.
Want to know more models available? See the full list of open models available in Fireworks AI on Foundry.
Enterprise Controls, Not Enterprise Overhead
The pitch isn't just performance. It's performance without the compliance and security trade-offs that slow down enterprise deployments.
Every model in Fireworks AI on Foundry runs within Azure's enterprise governance framework: unified access controls, audit logs, content filtering, and compliance tooling, through the same control plane you use for everything else. Foundry's built-in risk and safety evaluators let you validate model behavior in production before it becomes a problem.
The practical result: you can adopt a new frontier model, and keep adopting new ones as the landscape moves, without rebuilding pipelines, fragmenting tooling, or filing a new security review every time. All within your existing Azure commitments.
Join Us at Microsoft Build
Want to go deeper? Join us at Microsoft Build 2026to see Fireworks AI on Foundry in action—learn how enterprise teams are using the latest open models to build production-grade AI applications and talk directly with the teams behind the platform.
Microsoft