Anthropic launches Message Batches API in beta, for batches of up to 10K queries that are processed within 24 hours and cost 50% less than standard API calls
a cost-effective way to process vast amounts of queries asynchronously. You can submit batches of up to 10,000 queries at a time. Each batch is processed within 24 hours and costs 50% less than standard API calls. https://www.anthropic.com/... Shaun Ralston / @shaunralston : Good news, devs! @Anthropic just joined @OpenAI with batch API pricing; 50% cheaper than standard API calls. Lower costs on best models = a win for everyone! [image] Shreya Rajpal / @shreyar : Feature convergence between OpenAI and Anthropic Both now offer: - Batch Mode - Function Calling - Artifacts / Canvas Overall win for developers with more interop across providers @anthropicai : The Batches API is available today in beta. Requests made to the Batches API do not count against your standard API rate limits. Get started with the docs: https://docs.anthropic.com/... Alex Albert / @alexalbert__ : In my opinion, any task that doesn't need to be processed in real-time should be sent through the Batches API. Not only does it save you money, but it eliminates all the complexity of running many parallel queries, dealing with rate limiting, etc.
Context & Ripple Effects
Anthropic had already focused its API roadmap on reducing repeated-work costs, including prompt caching for frequently reused context. Message Batches extends that efficiency push to work that can wait, rather than changing the model’s standard real-time API behavior.
The launch also places Anthropic in an API market where OpenAI had established business-facing access with dedicated-capacity options through its ChatGPT API for businesses. The immediate distinction is not a new model capability, but a lower-priced processing lane tied to a 24-hour completion window.
First-order effects
- Developers can submit up to 10,000 asynchronous queries at once at half the standard API-call cost, making deferred bulk jobs cheaper to run on Anthropic.
- Batch requests are separated from standard rate limits, so teams can move qualifying high-volume work off their interactive API traffic.
Second-order effects
- Application teams gain an incentive to split workloads by urgency: latency-sensitive tasks remain on standard calls, while back-office evaluation, classification, or other deferrable jobs can use batches.
- The offering strengthens pressure on competing model APIs to distinguish pricing by turnaround time, not only by model or token volume.
Third-order effects
- If providers continue offering cheaper delayed execution, inference pricing is likely to become more explicitly capacity-aware: customers trade latency for lower unit costs while providers steer flexible demand toward available compute.
- That could make workload orchestration a more consequential part of AI application design, alongside techniques such as caching repeated prompt context; the benefit depends on a customer’s ability to tolerate delayed results.
The trend: Frontier-model APIs are increasingly segmenting inference by service level, using flexible latency to lower costs and better allocate compute capacity.