LinkedIn is using OpenAI's models to train smaller, less expensive alternatives that can handle recruiting tasks at up to 67% lower computing cost, a approach that points to a way companies can reduce their reliance on costly frontier models.
Enterprise AI · LinkedIn and OpenAI models
Smaller models, up to 67% lower compute cost for recruiting tasks
LinkedIn is using OpenAI's models to train compact, specialized alternatives for recruiting work, a shift that points to a way companies can rely less on costly frontier systems.
67%
Upper-bound reduction in computing cost for the smaller models
1/3
Remaining cost share at the 67% upper bound (about 33 of every 100 cost units)
Recruiting
Target domain for compact, function-specific models
Relative computing cost per task
Each block = 10% of the frontier-model cost. Savings shown at the 67% upper bound; actual savings will vary by workload.
100%
Frontier model
~33%
Smaller model
How the approach works
1. Frontier models
→
2. Compact models
→
3. Routine recruiting tasks
→
4. Lower cost at scale
The central question
Savings only matter if the smaller models stay reliable for real hiring decisions.
Which OpenAI models were used is not detailed
How performance was measured against the original systems is not detailed
Whether accuracy and other quality metrics match the larger models is not confirmed
Hiring carries legal and fairness obligations in many jurisdictions
Promising
Cuts the cost of running high-volume language tasks
Offers a template for customer service, sales and HR
Reserves large models for harder problems
Open questions
67% is an upper bound, so typical savings will vary
Unclear how much savings come from size reduction versus other optimizations
Quality and fairness documentation at production scale remains to be seen
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...