Together AI launches guaranteed inference for open models with 99% SLA

Together AI on July 8, 2026 presented Provisioned Throughput — a new inference format with reserved capacity for open models, guaranteed 99% uptime SLA and token-based pricing, similar to closed APIs.

What is Provisioned Throughput and how does it work?

Until now, companies working with open-weight models chose between two options. The first — serverless inference: convenient and "pay-as-you-go," but without guaranteed throughput, making it unreliable for production workloads. The second — dedicated GPUs: predictable and with full control, but requires calculations in GPU-hours, capacity planning and inference stack management. Provisioned Throughput is the third model.

A client purchases PTU (Provisioned Throughput Units) — a fixed share of capacity: guaranteed number of tokens per minute for a specific model, reserved exclusively for them. Input, cached, and output tokens deduct PTU with different coefficients, allowing cost optimization for actual traffic. The API for Provisioned Throughput is identical to the API for Serverless and Dedicated Inference — you can switch without changing code.

Key product parameters:

• Launch — July 8, 2026, minimum contract term — one month
• Available models at launch — MiniMax M3 and GLM-5.2
• Regions — North America, EMEA and others
• SLA — 99% uptime
• Price — $0.05 per PTU per minute
• Savings — up to 90% compared to Claude Opus 4.8 according to public price lists

Why open models are displacing closed APIs

Together AI records sharp growth in demand for open models: over nine months, token volume through the company's API grew from 30 billion to over 400 trillion per month. A significant portion of this traffic represents workloads previously executed through closed APIs of major labs.

Companies that switched to open models report 6–20x reduction in inference costs. The main reason — the ability to distribute tasks along the "Pareto boundary": complex tasks go to powerful models, routine ones — to lighter and cheaper ones. This flexibility is exactly what's needed to scale AI agents: when each employee runs several parallel agents for coding, financial calculations and automation — the token budget becomes a line item decided at the board level.

Provisioned Throughput addresses exactly this: predictable capacity, clear pricing and SLA — without needing to hire a DevOps team to manage GPU clusters.

What this means

Provisioned Throughput closes a gap that long existed in the open models market: predictable capacity and SLA while maintaining token-based pricing simplicity. For AI teams scaling agent workloads, this is a real path from closed APIs to open models without sacrificing reliability.

Frequently Asked Questions

What models are available in

Provisioned Throughput right now?

As of the July 8, 2026 launch, MiniMax M3 and GLM-5.2 are available. Together AI announced expansion to 40+ models selected for production use — timelines not specified.

What is the minimum contract cost?

The minimum term is one month. Cost is determined by the number of PTUs: $0.05 per PTU per minute. Each PTU guarantees fixed throughput in tokens per minute for the selected model — with no additional infrastructure costs.