%20rotate(-180)'%20xlink:href='%23linear-gradient'%3e%3c/linearGradient%3e%3clinearGradient%20id='linear-gradient3'%20x1='841.45'%20y1='247.11'%20x2='3086.34'%20y2='247.11'%20gradientTransform='translate(4479.33%203591.33)%20rotate(-180)'%20xlink:href='%23linear-gradient'%3e%3c/linearGradient%3e%3c/defs%3e%3cg%20id='Z'%3e%3cpath%20id='_x31_'%20fill='url(%23linear-gradient)'%20d='M2695.1,484.18c20.1,19.58,37.87,42.03,52.67,67.11,57.79,97.91,59.52,215.79,4.58,315.34l-992.94,1799.47h200.21c274.56,0,527.42-149.27,660.07-389.66l989.02-1792.26h-913.62Z'%3e%3c/path%3e%3cpath%20id='_x32_'%20fill='url(%23linear-gradient1)'%20d='M2674.21,823.5h0c84.15-152.48-26.15-339.33-200.31-339.33h-1275.39c-274.56,0-527.42,149.27-660.07,389.66l-80.32,145.55h2108l108.09-195.88Z'%3e%3c/path%3e%3cpath%20id='_x33_'%20fill='url(%23linear-gradient2)'%20d='M1400.9,3611.82c-20.1-19.58-37.87-42.03-52.67-67.11-57.79-97.91-59.52-215.79-4.58-315.34l992.94-1799.47h-200.21c-274.56,0-527.42,149.27-660.07,389.66l-989.02,1792.26h913.62Z'%3e%3c/path%3e%3cpath%20fill='url(%23linear-gradient3)'%20d='M1421.79,3272.5h0c-84.15,152.48,26.15,339.33,200.31,339.33h1275.39c274.56,0,527.42-149.27,660.07-389.66l80.32-145.55H1529.88l-108.09,195.88Z'%3e%3c/path%3e%3c/g%3e%3c/svg%3e)
On June 30 Anthropic released Claude Sonnet 5, describing it as “built to be the most agentic Sonnet model yet.” For a bank, the claim that matters is the price tag more than the benchmark chart. Sonnet 5 lists at $3 per million input tokens and $15 per million output, with introductory pricing of $2/$10 through August 31, while Anthropic says its “performance is close to that of Opus 4.8, but at lower prices.”
Near-flagship capability at mid-tier prices changes vendor economics overnight. Every software vendor with an AI feature now has a strong financial incentive to swap the model underneath it. Most vendor contracts let them do exactly that without telling you.
Why a cheaper model is a compliance event
A model swap inside a vendor product changes behavior you may have validated, such as alert-triage thresholds, document summaries and customer-facing language. The May consent order against Community Federal Savings Bank showed what a miscalibrated automated triage system costs. A silent model change is how a calibrated system becomes a miscalibrated one without anyone deciding anything.
Falling frontier prices are a feature for your budget and a bug for your change management. The cheaper the swap, the more often it happens.
- Ask every AI-bearing vendor which model and version is in production for your instance, and log the answer. Repeat the question quarterly.
- Put model-change notification into new and renewing vendor contracts. A swap that changes output behavior is a change-management event.
- Re-run your validation samples after any disclosed model change, and date the results in the vendor file.
Zovos keeps the model-version log for every AI vendor in your inventory and flags the ones that have gone a quarter without an attestation. The record manages the swap risk, so you are not relying on trust.
This is for information only and is not legal or investment advice. Verify all figures against the linked primary sources before acting.