New playbook: when your own AI model pays. Read
Clay (sales data software, ~300 employees) reported this on 27 Oct 2026 in vendor case with a named customer (source). It belongs to the learning loop layer.
A post-trained Qwen 3.6 35B-A3B router that predicts answer quality per query and sends each Claygent research query to the cheapest model that will answer it well.
Inference cost down up to 47% at retained quality; 48.2% savings at the 92.5% quality point vs always using the frontier model; can route hundreds of millions of queries a month in a fraction of a second per decision.
The number covers claygent, one of Clay's highest-volume workloads; offline evaluation plus a live A/B test on production traffic.
“Applied Compute cut our Claygent inference costs by up to 47% while maintaining quality,” Clay said.
Reviewed corrections are kept as test cases and later used to train models.
Read the guide to the learning loop.
96%
lower estimated serving cost for one agent
A daily continual-learning loop for Sidekick's GraphQL agent: an expert-calibrated LLM judge scores production conversations, a self-healing pipeline repairs failures, and the repaired trajectories are trained into a smaller model (SFT, then GRPO against the judge) that replaces the frontier model (source).
13.8x
lower cost per task, with higher accuracy than the best frontier model
A Qwen3-235B model fine-tuned on Tinker with investor-labelled data to do six daily information-filtering tasks (e.g. which news matters to a macro investor) (source).
1% → 99%
valid structured output after fine-tuning a small model
A proof of value on Red Hat OpenShift AI: GPT-4.1 generated synthetic training data, then small models (Phi-4 Mini 3.8B, Qwen 2.5 1.5B/7B) were fine-tuned for multilingual structured extraction from social media, deployable locally (source).