SteamDB vs SteamSpy vs VG Insights vs Gamalytic (2026)

Here is the short version: SteamDB has the most accurate Steam data on the internet and publishes zero sales estimates. SteamSpy invented the sales-estimation genre and has been running on a degraded methodology since Steam’s April 2018 privacy change. If you want an actual units-and-revenue number for a specific game in 2026, the two tools doing that job seriously are VG Insights (now owned by Sensor Tower) and Gamalytic, and Gamalytic is the only one of the four that publishes a detailed error study on its own estimates.

The rest of this post covers where each tool’s numbers come from, what accuracy each vendor claims (self-graded exams, all of them), and which tool to open for which job. I verified everything against the tools’ own documentation in early July 2026.

SteamDB reports facts. The other three report estimates. Mixing the two up is the most common data mistake I see indie devs make, and it runs through everything below.

SteamDB is a third-party tracker (not affiliated with Valve) that records what Steam’s own systems expose publicly. Its FAQ is explicit about what that means: concurrent player counts come directly from the Steam API, are exact rather than estimated, and update every few minutes. On top of that you get full price history in every regional currency, follower counts over time, tag and metadata changes, and depot-level patch history.

What SteamDB does not have, anywhere, is a sales or revenue estimate. This surprises people because journalists cite SteamDB constantly, but the FAQ itself points readers to SteamSpy, PlayTracker, VG Insights, and Gamalytic for owner estimations. SteamDB’s position is that it publishes data, not guesses.

That restraint is exactly why it stays useful. Because the concurrent player (CCU) numbers are exact, you can build your own estimates on top of them with known math. A launch-week peak CCU multiplied by a genre-appropriate factor gets you into the right order of magnitude for units sold - we walk through the full method in how to estimate Steam sales from SteamDB, and the CCU-specific math has its own guide in estimating Steam sales from concurrent players.

SteamSpy, launched by Sergey Galyonkin in 2015, worked by sampling millions of public Steam profiles and counting who owned what. It was never exact, but the sampling method was statistically honest and the error bars were quantifiable.

In April 2018, Valve changed Steam’s privacy defaults so that game libraries are hidden unless a user opts in to sharing them. That single change removed SteamSpy’s data source overnight. The site came back a few weeks later on a machine-learning model that infers ownership from indirect signals, and Galyonkin himself described the rebuilt version as not very accurate. Nothing has structurally improved since; SteamSpy’s own about page still warns that figures are extrapolated from limited samples and that the site is completely unreliable for recently released games.

In practice, here is what SteamSpy gives you in 2026:

• •Owner counts as wide buckets (“20,000 .. 50,000”), not point estimates. A 2.5x-wide range is honest about uncertainty but useless for revenue math.
• •Genre and tag aggregates that are still directionally interesting for market sizing, because errors partially cancel out across thousands of games.
• •A free API that a lot of academic papers and hobby projects still build on, mostly because it is free and stable, not because it is current.

Extra features sit behind Patreon pledges rather than a normal subscription. I keep SteamSpy in the list because you will run into its numbers constantly in old articles and Reddit threads, and you need to know how much salt to apply: a lot. For any game released after 2018, prefer the two tools below.

VG Insights (VGI) was founded in 2020 by Karl Kontus and built its estimates on an upgraded version of the Boxleiter method - the old “reviews times a multiplier” trick, which we break down in the Boxleiter method explained. Instead of one global multiplier, VGI’s methodology article describes fitting review-to-sales ratios against a database of over 10,000 games with known sales figures, then adjusting per game for price, genre, age, and review volume, and sanity-checking against Steam’s weekly top sellers charts.

Their stated accuracy claim: estimates land within about 5% at an aggregate level. That is careful wording, because aggregate accuracy says little about any single game. Individual games can and do miss by much more; the review multiplier alone spans a 20-60x range depending on genre.

The big 2025 news: Sensor Tower acquired VG Insights in March 2025, and the product now runs as “Video Game Insights by Sensor Tower” covering 150,000+ Steam titles plus PlayStation and Xbox data. The vginsights.com URLs now redirect into Sensor Tower’s platform, which is why half the methodology links you find in older blog posts are dead. The founding team’s public write-up of the approach survives in their Game Developer article on estimating Steam sales.

On price: VGI historically ran free indie tools plus a cheap indie subscription (around $15/month via Patreon in its early years). Post-acquisition, the free Steam analytics tools still exist, but current paid pricing sits inside Sensor Tower’s sales-led platform and is not published as a simple price list I could verify. If a predictable monthly bill matters to you, that opacity is itself a data point.

Gamalytic is the youngest of the four and the one that has leaned hardest into transparency. Its methodology, described across its about page and blog, combines several independent signals instead of relying on reviews alone: tracked top-seller rank movements, concurrent player counts, a tuned Boxleiter-style review model, and regression across games with publicly confirmed sales.

The accuracy claim is unusually specific. In Gamalytic’s self-published tests against games with known sales figures, 77% of individual game estimates landed within a 30% margin of error, 98% within 50%, and aggregate-level estimates were about 99% accurate.

Read that chart with both eyes open. It is a vendor grading its own homework, the test set skews toward games popular enough to have announced their numbers, and Gamalytic itself flags that estimates degrade for small games, free-to-play titles, and games sold heavily in bundles. Still, no competitor publishes anything this concrete, and “77% within 30%” is a genuinely useful error bar for planning: it means any single Gamalytic number should be read as a range, roughly 0.7x to 1.3x.

Pricing is published and indie-friendly. As of the March 2025 pricing update: a free tier with core per-game estimates, a Starter plan at $25/month (all filters, a year of historical data, wishlist estimates, and 2,500 API requests per day), and a Pro plan at $75/month for deeper history and export. Verify the current numbers on their pricing page before subscribing; this is exactly the kind of detail that changes.

Point-in-time comparisons are rare because developers seldom announce exact figures, but here is one documented case. In September 2023 the developers of Dwarves: Glory, Death and Loot announced they had sold over 30,000 copies. Gamalytic published a comparison of what each estimation engine showed at that moment:

Full provenance warning: the comparison came from Gamalytic’s own account, so of course it is a case where Gamalytic looked best. I include it anyway because the SteamSpy number is the real lesson - off by 5x in the wrong direction, on a post-2018 indie release, which is exactly the failure mode the privacy change created. VGI undershooting by 40% on a single mid-size game is also consistent with its “accurate in aggregate, noisy per game” framing. One game proves nothing about averages; it does show the spread you should expect when you cross-check.

That cross-check habit is the actual takeaway. When I research a comparable game, I pull both VGI and Gamalytic, and if the two disagree by more than 2x I treat the number as unknown and fall back to review counts and CCU math.

And the decision rules I actually follow:

1. 01Never quote an estimate without naming the tool. “Gamalytic estimates 40k units” is a claim you can check; “it sold 40k” is not.
2. 02Facts before estimates. If SteamDB’s exact CCU contradicts an estimator’s story, believe SteamDB and re-derive.
3. 03Two estimators or zero. Within 2x of each other, average them. Further apart, do your own review-multiplier math.
4. 04Apply the error bar. A single Gamalytic figure is a 0.7x-1.3x range at best; budget against the low end.

Steam Page Analyzer is not a fifth estimation database and does not pretend to be. Our free revenue calculator does the downstream math the four tools above leave to you: give it a review count or a units estimate from VGI or Gamalytic, and it applies the current review-to-sales multiplier, refund rates, regional pricing mix, and Steam’s 30% cut to turn that into net developer revenue. It is the “so what does that mean in dollars” step. We compared how the popular calculators handle that math, and where they quietly disagree, in our Steam revenue calculator comparison.

The workflow that has held up best for me: SteamDB for the facts, Gamalytic plus VG Insights for the estimate, our calculator for the net-revenue translation, and the Steam Page Analyzer itself for the question the databases cannot answer - whether your own store page will convert the traffic those comparable games are telling you to expect.