PeriFlow: Keep production agent loops fast and reliable
What is PeriFlow?
FriendliAI provides inference for open-weight and frontier models, aimed at production AI agents rather than demos. It offers Model APIs for per-token access to models such as GLM-5.3, GLM-5.3-Flash, DeepSeek and Qwen, dedicated endpoints on A100 through B300 GPUs, and a BYOG option for running the Friendli inference stack on your own hardware. The platform emphasises low-latency token streaming, long-context stability, and consistent tool calling as traffic scales. The company says its inference stack combines continuous batching, custom kernels, speculative decoding, KV cache optimisation and cache-aware routing with automatic failover and autoscaling. Published figures include roughly 7x faster output token speed, about 90% lower inference cost and a 99.99% uptime SLA. Model API pricing is listed per million tokens, dedicated GPU capacity is billed per second, and enterprise capabilities such as VPC or on-prem deployment are arranged by contract.
Focus areas: Model API, Model Inference.
Who is PeriFlow for?
- Engineering teams running production AI agents
- Platform teams serving open-weight models at scale
- Companies needing dedicated GPU inference or BYOG deployment
How much does PeriFlow cost?
Pricing model: paid. Always check the vendor’s own pricing page before you buy.
| Plan | Price | What you get |
|---|---|---|
| Model APIs | Pay per token | Text and vision models billed per 1M tokens; STT billed per audio minute |
| Dedicated Endpoints | $4.0/hour | On-demand GPU deployment billed per second; A100 80GB $4.0/hr, H100 80GB $5.0/hr, H200 141GB $7.0/hr, B200 180GB $9.0/hr, B300 288GB $12.0/hr |
Is PeriFlow worth it?
A usage-priced inference platform for teams that need open-weight models to hold up under production agent traffic
Where PeriFlow is strong
- Usage-based pricing across per-token APIs and per-second GPU billing
- Serving features aimed at long-context and tool-calling workloads
- Deployment options spanning hosted APIs, dedicated GPUs and your own hardware
Where it falls short
- Model API pricing is only published for a subset of listed models
- Enterprise features and BYOG pricing require contacting sales
How do you get started with PeriFlow?
- 01Create an account and issue an API key
- 02Pick a model from the Model APIs list or provision a dedicated endpoint
- 03Send requests through the API and monitor usage and latency
PeriFlow: frequently asked questions
How do I get started with FriendliAI?
The pricing page describes instant self-serve access: you sign up, issue an API key and call the Model APIs, or contact the team for dedicated and enterprise deployments.
How is FriendliAI priced?
Model APIs are billed per token, with published rates per 1M tokens for models such as GLM-5.3 and GLM-5.3-Flash, plus speech-to-text at $0.0015 per audio minute. Dedicated endpoints are billed per second by GPU type, from $4.0/hour for an A100 80GB to $12.0/hour for a B300 288GB.
Do I pay for idle GPU time on dedicated endpoints?
The pricing page states that on-demand deployment bills only for the compute you use, down to the second, with no extra charges for start-up times.
Can I run FriendliAI on my own GPUs?
Yes. The BYOG option runs the Friendli inference stack on your own GPUs, with pricing tailored to your infrastructure and deployment needs through a conversation with the team.
PeriFlow alternatives
The closest tools in our directory, matched on category, focus areas and audience.
| Tool | Best for | Free plan | Starting price | Focus areas |
|---|---|---|---|---|
| PeriFlow | Teams serving open-weight models to production agents | No | — | Model API, Model Inference |
| Chutes.ai | Developers running open-source model inference at scale | No | — | Model API, Model Inference |
| Samaira AI | Samaira AI is an AI inference platform offering scalable and secure deployment of machine learning models, simplifying the process for… | Unknown | — | Model Inference |
| Upstage AI | Teams automating document parsing and extraction with LLM APIs | No | $100/mo | — |
| Cohere | Enterprises needing private, controllable generative AI and retrieval | No | — | — |
| Not Diamond | Teams cutting LLM inference cost on coding agents | No | — | — |
| Novita | Developers running model inference, GPU jobs and agent workloads | No | — | Model API |
| Cerebras | An AI company that builds wafer-scale chips and AI supercomputers designed for training the largest neural networks. | No | — | — |
| Reka AI | An AI research and product company building powerful, multimodal large language models for the enterprise. | No | — | — |
| Mistral AI | Teams needing controllable, self-hostable language models | Yes | $14.99/mo | — |
| Baseten | Teams serving open-source or custom models at production scale | Yes | — | Model Inference |

