---
title: "Open, convenient and predictable: Introducing Provisioned Throughput"
description: "Companies adopting open-weight models for production workloads have faced a trade-off between convenient but best-effort serverless inference and powerful but complex dedicated GPU instances. Together AI’s Provisioned Throughput fills that gap with reserved inference capacity sold as Provisioned Throughput Units (PTUs). Each PTU is priced at $0.05 per minute and guarantees a fixed rate of tokens per minute for a specific model, with a 99% uptime SLA. Input, cached input, and output tokens draw down capacity at different rates, letting teams optimize for their traffic shape. Available now for MiniMax M3 and GLM-5.2 with capacity in North America and EMEA and a one-month minimum term, at full utilization costs can run roughly $0.36 per million input tokens and $2.16 per million output tokens, up to 90% below proprietary alternatives."
---

# Open, convenient and predictable: Introducing Provisioned Throughput

[Together AI](https://yomu.fyi/company/together-ai) · Jul 8, 2026

**Type:** Announcement

## Summary

Companies adopting open-weight models for production workloads have faced a trade-off between convenient but best-effort serverless inference and powerful but complex dedicated GPU instances. Together AI’s Provisioned Throughput fills that gap with reserved inference capacity sold as Provisioned Throughput Units (PTUs). Each PTU is priced at $0.05 per minute and guarantees a fixed rate of tokens per minute for a specific model, with a 99% uptime SLA. Input, cached input, and output tokens draw down capacity at different rates, letting teams optimize for their traffic shape. Available now for MiniMax M3 and GLM-5.2 with capacity in North America and EMEA and a one-month minimum term, at full utilization costs can run roughly $0.36 per million input tokens and $2.16 per million output tokens, up to 90% below proprietary alternatives.

## Context

Companies building production workloads on open-weight models previously had to choose between best-effort serverless inference and complex dedicated GPU instances. Together AI identified a need for reserved capacity with token-based pricing and availability guarantees, especially as token volume through its APIs grew from 30 billion to over 400 trillion tokens per month.

## Approach / What changed

Provisioned Throughput sells capacity as Provisioned Throughput Units (PTUs) at $0.05 per PTU per minute, each guaranteeing a fixed token-per-minute rate for a chosen model. Input, cached input, and output tokens draw down PTUs at different rates, enabling cost optimization based on traffic pattern. The service includes a 99% uptime SLA and a one-month minimum term, with identical API to serverless and dedicated inference.

## Takeaways

- Provisioned Throughput offers reserved inference capacity for open models with token-based pricing and a 99% uptime SLA, eliminating the need to manage GPU infrastructure.
- On MiniMax M3, one PTU provides 138,840 input tokens, 694,200 cached input tokens, or 23,140 output tokens per minute, with different consumption rates for each token type.
- Costs at full utilization on MiniMax M3 are approximately $0.36 per million input tokens and $2.16 per million output tokens, up to 90% less than proprietary APIs like Claude Opus 4.8.

## FAQ

### What models are supported at launch?

Provisioned Throughput is available today for MiniMax M3 and GLM-5.2, with more models coming. Capacity is offered in North America, EMEA, and additional regions.

### How do PTUs account for different token types?

A PTU covers input, cached input, and output tokens, each drawing down capacity at different rates. On MiniMax M3, one PTU delivers 138,840 input tokens, 694,200 cached input tokens, or 23,140 output tokens per minute.

### How does Provisioned Throughput compare to serverless and dedicated inference?

Serverless is best-effort for development, dedicated gives full control for custom models, while Provisioned Throughput offers reserved capacity with a 99% SLA for production workloads. All three share the same API, making migration straightforward.

**Tags:** [Inference](https://yomu.fyi/topic/inference), [LLMs](https://yomu.fyi/topic/llm), [Open Source](https://yomu.fyi/topic/open-source), [Provisioned Throughput](https://yomu.fyi/topic/provisioned-throughput)

- Source: [Together AI](https://www.together.ai/blog/provisioned-throughput)
- Source URL: https://www.together.ai/blog/provisioned-throughput
- Ingested by Yomu: 2026-09-19T10:13:39.830Z

[Read original post](https://www.together.ai/blog/provisioned-throughput)