OpenAI drops a 50% off Batch API and supercharges Assistants

If you are trying to ship AI into production inside a large company, the biggest blocker is rarely model smarts. It is almost always security compliance and runaway cloud bills.

OpenAI just rolled out a major stack of enterprise-focused upgrades for its API customers, targeting data privacy, access controls, developer tools, and infrastructure costs.

Tighter fences and team-level guardrails

For companies running strict security architectures, OpenAI introduced Private Link. It establishes direct communication between Azure environments and OpenAI, minimizing exposure to the open internet.

Native Multi-Factor Authentication (MFA) is also live, helping organizations meet tightening compliance rules.

Managing multiple engineering teams gets much easier with the new Projects feature. Project owners can now scope roles, restrict API keys to designated workspaces, and set hard usage caps to prevent surprise bills.

Teams can also create service account API keys, meaning critical production scripts are no longer tied to individual employee credentials.

Assistants API gets 500x more files and a 50% discount

The Assistants API received a substantial overhaul. The headline feature is an upgraded file search capability that can ingest up to 10,000 files per assistant—up from just 20 previously.

It features parallel multi-threaded search, improved reranking, and new vector_store objects that handle parsing, chunking, and embeddings automatically across different assistants.

Developers finally get streaming support for real-time conversational responses, tighter token usage limits per run, and support for fine-tuned GPT-3.5 Turbo models.

On the cost front, OpenAI launched the Batch API. Developers can queue non-urgent jobs like model evaluations, summarization, and data generation with a 24-hour turnaround window at a flat 50% discount off standard shared rates.

For massive workloads with steady token throughput on GPT-4 or GPT-4 Turbo, committed throughput deals now offer discounts between 10% and 50%.

Why it matters

We think this release signals OpenAI's aggressive transition from developer sandbox to full-blown enterprise utility.

For developers building knowledge retrieval apps, bumping the document ceiling from 20 to 10,000 files while automating chunking and embeddings effectively wipes out the need to manage external RAG plumbing for basic use cases. If you run nightly batch jobs or bulk evaluations, switching to the Batch API immediately cuts your token burn in half.

By checking off Private Link, MFA, and service accounts, OpenAI removes the classic security and legal objections that enterprise IT departments throw at cloud AI APIs.

However, baking document storage and retrieval directly into the API creates deep architectural lock-in. It is convenient today, but walking away from OpenAI later will mean rewriting both your orchestration logic and your vector data pipeline.