LinkedIn is using OpenAI's models to train smaller, less expensive AI systems that can handle recruiting tasks at up to 67% lower computing cost, a move that points to a way companies can reduce their reliance on costly large models.
LinkedIn · OpenAI models
Smaller models, lower bills: LinkedIn cuts recruiting AI costs by up to 67%
LinkedIn is using OpenAI's models to train compact AI systems that handle specific recruiting tasks at a fraction of the computing cost of running larger models, pointing to a way companies can rely less on expensive frontier systems.
Computing cost, relative to the larger model
Each block = 10% of baseline cost. Maximum reported saving shown.
Up to 67% lower cost means the smaller model runs at about one-third of the baseline in the best case.
67%
Maximum reported reduction in computing cost
Narrow
Smaller models target specific recruiting functions
Premium
Large models retained for complex tasks
How the approach works
Step 1
Large frontier model generates data or guides training
→
Step 2
Compact model is trained on a specific task
→
Step 3
Deployed at high volume for screening, matching, and outreach
Read the number carefully
67% is a ceiling, not a guaranteed average
Which recruiting tasks are covered has not been fully detailed
Benchmarks against the original models are not fully published
Accuracy trade-offs accepted in exchange for savings are unclear
Why it matters
Inference costs can become a major line item once AI runs everyday business processes
Cost-optimized models open a path to broader deployment without abandoning frontier capabilities
A high-volume HR setting shows a routine-versus-premium model split is achievable
Open questions
How well smaller models handle edge cases
How far the savings extend beyond recruiting
Whether other large platforms adopt similar strategies
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...