Together AI raises Batch Inference API limits 3,000-fold and cuts prices by 50%

Together AI has updated its Batch Inference API, a service for processing large numbers of LLM requests with deferred execution. The company announced three major improvements: limits raised 3,000-fold, a 50% price cut, and a redesigned interface to simplify use.

Scaling without bottlenecks

The main change concerns limits. The previous ceiling was 10 million tokens per user per model; it is now 30 billion. This is not just a number: it solves an architectural problem faced by companies that process huge datasets.

Previously, teams handled large workloads the simple way: they split the dataset into pieces, created many small batch jobs, tracked each one separately and coordinated the results. That was inconvenient, slow and expensive. Now you can upload the entire dataset in a single operation and get results within a 24-hour SLA, often much faster.

Pricing was updated at the same time. Batch processing now costs roughly half as much as the real-time API for the same amount of compute. When you are talking about billions of tokens, the price difference becomes significant for project budgets.

Any model, a simple UI

The API now works with all 40+ models on the Together platform, including private deployments. Previously the choice was limited to a few models, which caused problems for teams that wanted to experiment with and test different models in batch mode.

The interface has been completely redesigned. Before, you had to write API calls, read the documentation and debug code. Now everything happens in a web app: creating a job, monitoring progress, downloading results. A few clicks and it's done. This lowers the barrier to entry for teams that don't want to spend time writing code for every batch request.

Who needs this

• Sentiment analysis and text classification across millions of documents
• Fraud detection: scanning millions of payments and transactions
• Synthetic data generation for training new models
• Vectorization of large text corpora (embedding generation)
• Content moderation on social networks and platforms with UGC
• Benchmark tests to evaluate and compare model quality

A concrete example: the lab Inception Labs already uses the batch API as the foundation of its production workflow. According to co-founder Vladimir Kuleshov:

"We rely on the Batch Inference API to process very large volumes of requests.

The high limits let us run gigantic experiments without bottlenecks. Jobs finish significantly faster than the 24-hour SLA, often within a few hours."

What this means for the industry

Batch inference is moving from a specialist niche to a mainstream tool. High costs and technical complexity used to be a serious barrier. Only large research labs, government projects and big corporations could afford batch processing. Now startups and mid-sized teams have access to the same tooling.

The 50% price cut and the 3,000-fold limit increase remove the main obstacles to wide adoption. A surge in batch inference use in production applications is expected in 2025, from real-time-scale content moderation to synthesizing large volumes of training data for fine-tuning custom models.