New playbook: when your own AI model pays. Read
Factory (software, <200 employees) reported this on 1 Jul 2026 in company self-report (source). It belongs to the learning loop layer.
Two small LoRA adapters on Qwen 3.6 35B-A3B that decide whether a code commit leaks a secret (one flags risk, one clears false alarms) inside its Droid Shield commit gate.
At a strict 5% false-alarm budget, recall 0.698 vs 0.588 for GPT-5.5 and 0.574 for Opus 4.8 (non-overlapping 95% CIs), at lower cost and latency.
The number covers 427-example evaluation built from Samsung's CredData benchmark.
“…they are the strongest or tied-strongest classifiers on both tasks compared to the frontier, at a fraction of the cost,” Factory said.
Reviewed corrections are kept as test cases and later used to train models.
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).