Carvana is the first U.S. auto ABS issuer to report both scores on essentially every loan. The two often disagree on the same borrower, both show up in price, and early performance can't yet tell them apart.

Ask which credit score an auto lender "uses," and the honest answer is usually: its own. CCM read the underwriting sections, pool selection rules and loan-level data notes for the 25 SEC-registered auto loan ABS programs it tracks. Nearly every lender decides on a proprietary score (e.g. Toyota's TMCC score and GM Financial's Custom Score). FICO, VantageScore or both feed into it as one input among many.

What the bond market sees is different. FICO is the score on the loan tape, in the pool tables and, often, in the pool selection rules themselves. Toyota requires a FICO of at least 620 for a loan to enter a trust. Since mid-2024, CarMax has used a FICO of 650 to sort loans between its two securitization programs, and World Omni routes Toyota-brand loans with a FICO below 650 into separate WOSAT trusts. Of the $144 billion of active balance on these programs' August 2026 tapes, 85% reports a FICO and no VantageScore.

VantageScore has been moving into the decision process for years: Toyota's prospectuses have described it as an input to its scoring engine since at least 2021, and Exeter and Bridgecrest have since switched from FICO to VantageScore. Public disclosure lags: Toyota reports only FICO, and Exeter and Bridgecrest report a single unlabeled score field that mixes the two.

What Carvana changed. In December 2025, Carvana stopped using FICO when it pre-qualifies a customer. Its prospectuses now state that the originators "do not use FICO® scores in connection with their credit underwriting process." Offers come from Carvana's proprietary risk models, which incorporate third-party credit scores such as VantageScore, raw bureau data and alternative data. Carvana still pulls a FICO at origination, specifically FICO Auto Score 8, and since its first deal of 2026 it reports both scores for each loan, labeled: "VS 794, FICO 824."^[1]

That makes Carvana's 2026 tapes the first in U.S. auto ABS to carry both scores on essentially every loan: two scores on the same borrower, from the same moment, in a pool where VantageScore was available to Carvana's models at prequalification and FICO, by Carvana's account, was not. The two scores usually agree. The story is in the loans where they don't.

The landscape. CCM tracked 130,546 loans in Carvana's three 2026 prime deals, 2026-P1, P2 and P3. 99.4% carry both scores. 698 loans (0.5%) carry only a VantageScore, and 80 carry a FICO only or no score. Loans originated as recently as July 2026 are still 99.3% dual-scored, so FICO isn't leaving Carvana's pools.

The two scores aren't on the same scale. FICO Auto Score 8 runs from 250 to 900; VantageScore runs from 300 to 850. In these pools, 3.6% of borrowers score above 850 on FICO, as high as 899. A FICO 700 and a VantageScore 700 aren't the same statement about a borrower. The market treats them as interchangeable labels anyway, so the useful question is how often they disagree.

Ranked against each other, they broadly agree: the rank correlation is 0.90. Loan by loan, they often don't. 18% of borrowers have scores 50 or more points apart, and 23% sit 15 or more percentile points apart within the pool. Co-borrowers don't explain it, since single-borrower loans disagree just as much. At a 660 cutoff, 13% of borrowers land on opposite sides depending on the score; at 620, 9%.

Both scores show up in price. Hold either one fixed, and the other still moves the APR by a similar amount. That FICO matters at all is a mild surprise, since Carvana says it stopped using FICO at prequalification. Its information may reach the price through other bureau data in Carvana's models, or through steps after prequalification; the public data can't say which.

VantageScore-only borrowers look different. The 698 borrowers with no FICO differ from dual-scored borrowers with the same VantageScore, in ways consistent with thinner credit files. They borrow less ($22,000 vs. $29,000), rarely have a co-borrower (about 5% vs. 26%), have higher payment-to-income, and pay about 3.5 points more in APR on average, from 2.5 to 4.7 points depending on the VantageScore band.

The pools changed, too, so a before-and-after comparison would mix the score change with collateral changes: the 2026 pools have longer terms, larger loans, newer vehicles and more co-borrowers than the 2025 deals. The cleaner test is inside the 2026 pools, where every loan carries both scores.

Performance: too soon to tell. Carvana's prime loans rarely go late early. Through August, 822 of the 77,779 dual-scored loans in 2026-P1 and P2 had been 30+ days past due at least once. 2026-P1 has eight months of history, P2 five and P3 two. On that early read the two scores have similar ability to separate loans that went late from those that didn't (AUC 0.75 for FICO, 0.74 for VantageScore).^[2]   FICO's slight edge doesn't hold on single-borrower loans or on 2026-P1 alone.

There's also a built-in selection effect. These loans were already screened by Carvana's models, which use VantageScore and other bureau data, and borrowers screened out never appear on the tape. That probably tilts a head-to-head inside these pools against VantageScore, and a near-tie can't show the two scores are equally good across all applicants.

CCM read: Most of the market still speaks FICO to investors, whatever its own models listen to. Carvana is the first to publish, loan by loan, the score it underwrote with alongside the one the market expects. It's too early to say which predicts better, and these pools can answer only part of that question: the borrowers VantageScore turned away never show up. What they can show is whether the borrowers VantageScore approved and FICO would have flagged hold up. CCM will track it, and so should anyone who prices auto credit off a FICO.