Vintage Gibson Price Estimator

We train a gradient-boosted model (XGBoost, quantile regression) on 15,340 verified sales collected from Reverb, dealers, and auction houses between 2015 and 2026. The model predicts the 25th, 50th, and 75th percentile of sale price from dozens of signals extracted from each listing.

On held-out sales the median estimate lands within 26.3% of the true price on average, and the typical range is calibrated — roughly half of real sales fall inside it, by construction. Estimates for rare combinations lean on the model rather than direct comparables; the comp count shown under each estimate tells you which regime you're in.

Like the vintage Fender estimator, this isn't one model — it's six. Out-of-fold cross-validation testing found Les Paul, SG, and Melody Maker each do best on their own (pooling SG with Melody Maker was a wash, not an improvement), while ES-335/345/330/175 pool into one “ES Series” model, Firebird/Flying V/Explorer pool into “Modernist” (pooling dramatically helped the notoriously thin Flying V data), and the six flat-top acoustics pool together. Two of these results reversed the initial genre-based guess going in.

Les Paul and SG both get one more dimension the other groups don't: a Standard/Custom/ Junior/Special/Deluxe variant selector. An early training pass priced a “1959 Les Paul Excellent” at just $11,400 because the raw model field pools genuine Bursts in with far-cheaper Juniors and Specials — splitting variant out fixed most, but not all, of that gap (see the note above about 1958–60 Standards specifically). SG has the same problem at smaller scale (roughly 2-3x, not 30x+) — early-1960s Standard sells for noticeably more than Junior — and adding its own variant dimension improved SG's overall accuracy meaningfully, not just its Standard/Junior pricing.

Celebrity provenance is a multiplier, not an exclusion: guitars with documented prior ownership by a notable player used to be dropped from training entirely on the theory that they'd distort the price curve, but most provenance claims turn out to be minor and barely move price — only a handful of true superstar cases (a Beatle, a headline blues or rock guitarist) sell for many times the going rate. Excluding all of them wasted otherwise-good comps to guard against a few outliers; the toggle above applies the average effect instead.

Below: SHAP values for the Les Paul model — how much each feature moves the price prediction, per listing. Manufacture year dominates, followed by sale year, Custom variant, condition, and Goldtop finish; headstock/neck break and refinishing show the expected negative direction.

Sale prices are verified transaction data, not asking prices. Grid predictions are evaluated at a reference configuration; secondary adjustments are applied client-side. Nothing you type here is stored.