Claim
The average frontier AI model maintains its performance lead for only 41 days before a competing model overtakes it on benchmarks, giving customers a credible switching option and negotiating power on a six-week cycle.
Mechanism
Model release cadence now outpaces any enterprise procurement cycle. When a model loses its benchmark lead at day 41, an equivalent or superior capability is available at the same or lower cost. The pricing premium attached to top-of-benchmark status expires with the lead. Customers who know the cycle can time renegotiations or competitive evaluations to coincide with each regime change. Vendors cannot sustain capability-based pricing premiums when parity arrives every six weeks.
Conditions
Holds when: the AI application is model-substitutable and the customer can switch without large migration costs; multiple competing labs are releasing at similar pace.
Fails when: deep integration has made vendor switching costly regardless of model benchmarks; one lab sustains a large enough lead that 41 days extends materially.
Evidence
"The average frontier model maintains its crown for 41 days."
"Customers gain negotiating leverage every 41 days."
Tunguz cites Microsoft's MAI-Code-1-Flash matching Claude Haiku 4.5 on SWE-Bench Verified using 60% fewer tokens, and xAI's Grok 4.5 scoring 54 on the Intelligence Index at $0.31 per task, 60% cheaper than leading models at comparable scores. Each of these represents a regime change within the 41-day window.
Signals
- Vendors offering multi-year AI capability commitments without lock-in clauses
- Customers requesting model-agnostic SLAs rather than named-model SLAs
- Procurement teams benchmarking AI suppliers quarterly rather than annually
Counter-evidence
Benchmark performance is not always what customers buy. Workflow integration, trust, and support quality all extend effective switching costs beyond the 41-day technical reset. Enterprise deals with embedded teams (FDE) create institutional lock-in that survives benchmark reversals. The 41-day cycle applies to buyer pricing position at the model layer, not to the full vendor relationship.
Cross-references
- The AI deployment bottleneck has shifted from model capability to customer adoption, making embedded engineering teams the institutional moat: FDE investment is the vendor countermeasure that converts raw capability into institutional lock-in, extending effective tenure beyond the 41-day window.
- Every structural dynamic in AI points toward commodity pricing at the model layer, with value concentrating further up the stack: Evans documents the structural direction; Tunguz quantifies the current cycle time.
- AI products retain customers between mobile-game and social-network benchmarks, with workflow integration separating the top cohort from the rest: the 41-day window shapes model-layer bargaining; the colander insight covers what determines retention at the product layer.