Upstart originated $3.0 billion in unsecured personal loans in Q1 2026 — up 50% year-over-year. Super prime borrowers now make up 33% of the mix, up from 27% last quarter. Every personal loan cohort since Q1 2023 is outperforming 2-year Treasuries by at least 385 basis points. April originations hit $1.274 billion — up 51% year-over-year. And Upstart’s AI model has closed only 12.6% of the distance to a perfect credit model — meaning 87.4% of the accuracy improvement still lies ahead. Here’s what consumer lenders should take from the data.
Upstart publishes more granular real-time data on AI-driven personal loan originations than any other company in consumer lending. Between the monthly origination volume dashboard and the Q1 2026 investor presentation released May 5, there is enough data to build a precise picture of where AI underwriting is in the personal loan market, how credit is performing across vintages, and what the origination trajectory looks like heading into Q2. For consumer lenders evaluating their own underwriting posture, this is the most useful comparative dataset available outside of your own portfolio.
Upstart’s real-time origination volume dashboard shows April 2026 personal loan originations of $1.274 billion — a 51% increase year-over-year. At 26.4 effective origination days in April, that translates to $48.2 million per day. The month-over-month trajectory has been consistently upward since the trough of late 2023, accelerating sharply through 2025 and into 2026.
In Q1 2026, Upstart originated $3.0 billion in unsecured personal loans across 410,854 loans — up 50% year-over-year in dollars and up 73% in loan count from Q1 2025’s 237,201 loans. The loan count declined sequentially from Q4 2025’s 443,784 — reflecting the same tax refund demand suppression that OppFi disclosed, with higher refunds reducing near-term borrowing need in January and February. The dollar volume held up better than the count because average loan sizes increased as super prime borrowers entered the mix at higher balance points.
The conversion rate — the percentage of rate inquiries that convert to funded loans — was 18.5% in Q1 2026. This metric has been remarkably stable across the past year: 19.4% in Q2 2025, 19.4% in Q3 2025, 21.0% in Q4 2025. The 18.5% Q1 reading reflects the seasonal tax refund demand suppression rather than a credit tightening signal. At 91% full automation — meaning no human intervention by Upstart in the underwriting decision — the platform is operating at near-maximum efficiency.
The most strategically significant data point in the Q1 personal loan section is the borrower segment mix. Super prime borrowers — defined as FICO 720 and above — made up 33% of personal loan originations in Q1 2026, up from 27% in Q4 2025. Core borrowers (FICO below 720, excluding small dollar) and small dollar loans make up the remainder.
This is a deliberate portfolio construction shift. Upstart’s AI model is approving more super prime borrowers — not because the platform has abandoned its near-prime roots, but because the model can identify attractive risk-adjusted returns in the super prime segment that traditional lenders with blunter scoring tools might price less efficiently. Super prime borrowers at Upstart are not the same as super prime borrowers at JPMorgan. They are borrowers who FICO scores as super prime and whom Upstart’s AI model has further validated as lower-risk than their FICO score alone would indicate.
The mix shift has two direct implications for the credit performance data. First, it improves aggregate portfolio metrics mechanically — more super prime loans lower expected loss rates. Second, it signals that Upstart is finding institutional capital appetite for higher-quality paper easier to satisfy than near-prime paper in the current macro environment — a rational response to the cautious posture of ABS investors and forward flow partners in Q1 2026.
The most important data in the investor deck for lenders evaluating Upstart as a capital deployment channel is on slide 22. Upstart reports that the net annualized return from investing equally in all personal loan cohorts since Q1 2023 would represent a 6.5% premium over the 2-year Treasury. More specifically, every individual cohort since Q1 2023 — including cohorts originated in the inflationary 2023 environment, the 2024 rate-hold environment, and the early 2026 Iran war environment — is exceeding Treasuries by at least 385 basis points.
This is the data point that explains why Fortress and Centerbridge each committed over $1.2 billion in forward flow agreements in the past two weeks. Institutional capital buyers with $40-$50 billion in AUM are committing at scale because the realized return data across multiple vintages — including stress vintages — demonstrates consistent outperformance of the risk-free rate by a meaningful margin. The 385 basis point floor is not a projection. It is a realized return floor across every cohort in the dataset.
The Upstart Macro Index — the company’s proprietary measure of macroeconomic risk to consumer credit — stood at 1.38 as of April 30, 2026. A UMI above 1.0 indicates macro conditions are adding credit risk relative to the baseline. At 1.38, the index is within the range observed since late 2025 — meaning the Iran war energy shock has not pushed macro credit risk outside the envelope that Upstart’s models have been operating within since the end of last year. That is a meaningful statement about the model’s resilience in the current environment.
Upstart measures its AI model’s accuracy using what it calls the “inaccuracy gap” — the distance between the model’s current predictions and a theoretically perfect model. A score of 0% would mean perfect predictions; 100% means no predictive power at all, treating every applicant as equal risk.
In Q1 2026, Upstart’s personal loan model scored 87.4% — meaning it has closed 12.6% of the distance to a perfect model since Q2 2018’s 92.3% reading. The traditional credit model Upstart uses as its benchmark scores 95.4% — meaning it has closed only 4.6% of the same distance. Upstart’s model is meaningfully more accurate than the traditional model, but — and this is the strategic point management emphasizes — 87.4% of the potential improvement still lies ahead.
For consumer lenders evaluating AI underwriting partnerships or their own model development, this framing is instructive. The accuracy gains from AI underwriting are real and measurable — Upstart’s model has materially outpaced the traditional credit model over eight years of development. But the gains so far represent a fraction of what the theoretical ceiling allows. The implication is that AI underwriting in consumer lending is early-stage technology, not mature technology — and that the competitive advantage it confers will compound over time rather than plateau.
The origination trajectory into Q2 is accelerating. April’s $1.274 billion at $48.2 million per day, up 51% year-over-year, is the strongest monthly volume Upstart has reported since the platform’s 2021-2022 peak. The Q1 loan count dip relative to Q4 was seasonal — tax refund demand suppression — not a credit tightening signal. With the refund buffer now exhausted and gas price demand pressure returning, Q2 personal loan origination volume is likely to run above Q1’s $3.0 billion pace.
The 385 basis point return premium over Treasuries is the most important competitive data point in consumer lending right now. Institutional capital buyers are committing billions to Upstart paper because the realized return data supports it. For non-Upstart lenders competing for the same institutional ABS investor and forward flow capital, this creates a reference point: your paper needs to demonstrate comparable or superior risk-adjusted returns to attract capital at comparable cost. If your loss rates are running above what the Upstart cohort data implies for similar borrower segments, your cost of capital will reflect that.
The super prime mix shift is a playbook worth studying. Upstart growing super prime from 27% to 33% of originations in one quarter reflects a deliberate response to the macro environment — moving up the credit quality stack when near-prime loss rates are under pressure while retaining AI model advantage over traditional lenders in the super prime segment. Lenders with flexible underwriting posture can apply the same logic: when macro risk rises, find the segments where your model advantage is largest relative to competitive pricing, not just the segments where volume is easiest.
The 87.4% inaccuracy gap means the AI underwriting advantage will compound. Upstart today is meaningfully better than the traditional credit model. Upstart in 2028 — after three more years of model refinement on a dataset that is growing exponentially — will be more meaningfully better still. For lenders building or evaluating their own AI underwriting capabilities, the relevant question is not whether AI underwriting outperforms today. It clearly does. The question is what your relative position will be in three to five years if you do not invest in model development now.
The Lending Pulse delivers weekly intelligence for consumer lending executives — macro signals, regulatory updates, earnings decoded, and company moves. Start your free month — no credit card required.
Start free month →