Statsig Review 2026: Features, Pricing, and Ownership

Statsig is one of the most capable experimentation platforms on the market and, in 2026, also the one that demands the most careful read of who is actually in control of it. In under a year it was acquired by OpenAI and then had its brand and customer base taken over by Amplitude. None of that invalidates the product, which still publishes its prices, still ships a substantial free tier, and is technically strong, but it does change how much weight the “vendor stability” criterion deserves in your decision. This review covers what the tool does, what it charges, how much of your test fits inside the free allowance, and what those two announcements mean in practice. For the category-level picture, see our neutral comparison of CRO and A/B testing tools.

Disclosure before anything else: Donnu is an A/B testing tool and competes with Statsig on part of this scope. This piece is written to be useful even to a reader who ends up choosing Statsig.

What Statsig is, one line per piece

Piece	What it does	Who usually cares
Feature flags and configs	Turns functionality on and off per segment, without a new deploy	Engineering, inside the normal delivery flow
Experimentation	Server-side and client-side A/B testing tied to those flags	Product, measuring what just shipped
Product analytics	Funnels, retention, and metrics over the same events	Teams that decide from their own data
Session replay	Playback of real behavior in the interface	Qualitative diagnosis of what the number showed
Warehouse native	Computation that runs on the company’s own data warehouse	Organizations with centralized data governance

The honest read of that table: Statsig is dev-first by design. The natural path is to instrument events, create a flag, and turn that flag into an experiment. It is not a tool where a marketer edits a page in the browser and publishes a test without engineering, and that is the first qualifying question to answer before any other.

Pricing: the published table, in full

Numbers checked on the official pricing page on 14 August 2026:

Plan	Price	Events included	Session replays	Analytics retention	Flag checks
Developer	Free, no card	2 million/month	50,000/month	1 year	Unlimited
Pro	150 dollars/month	5 million/month, overage 0.05 dollars per 1,000	100,000/month	Unlimited	Unlimited
Enterprise	Quoted	Event-based or experiment-based contract, volume discounts	Quoted	Unlimited	Unlimited

Enterprise adds warehouse-native execution, data-out integrations, warehouse ingestion, SSO, role and team based access control, priority support, and HIPAA eligibility under a specific agreement.

Two things that table makes explicit:

1. The free tier is genuinely large. Two million events per month and 50,000 replays is not a demo allowance, it is enough volume for a small operation to actually run on.
2. The flag is not the billing unit. Flag and config checks appear as unlimited on every plan. Anyone comparing feature flag tools on cost-per-evaluation at high traffic should record that difference, which our feature flag tools comparison also covers.

Who controls the platform: the two announcements that matter

This is the section that makes this review worth more than reading the pricing page, and it is entirely factual.

On 2 September 2025, OpenAI announced its acquisition of Statsig in an all-stock deal reported at about 1.1 billion dollars, with founder and CEO Vijaye Raji becoming OpenAI’s CTO of Applications and the Statsig team joining him.

On 5 May 2026, Amplitude announced it is taking over the Statsig brand and customers. In the words of the official announcement, Amplitude will maintain and develop the existing Statsig platform, both cloud and warehouse-native, including support for existing customers, and will work closely with the Statsig team at OpenAI through the transition, starting to build a more integrated roadmap across the two platforms.

The neutral read: continuity committed in writing is not the same as continuity guaranteed, and it is also not a sign of abandonment. Two practical consequences, both defensible regardless of your opinion about the deal:

• Prefer shorter contracts while the transition is in progress, and treat renewal as a decision point rather than a formality.
• Keep your experiment data exportable. That holds for any vendor, and it holds more when the platform has just changed hands.

How much of your test fits in the free allowance

Per-event pricing ties the cost of the tool to the volume of your instrumentation, not to the number of experiments you run. Set the calculator to your own baseline rate:

Current conversion rate%Minimum detectable effectrelative (%)absolute (pp)Confidence909599%Power8090%Visitors per week (total)Testtwo-sidedone-sided

-Visitors per variation

-Total (2 variations)

-Estimated duration

Two-proportion normal approximation, 2 variations (50/50). Tweak the inputs and watch it update live.

The worked example

A SaaS product converts 4% of the visitors entering its signup flow and gets 40,000 users per week into that flow. To detect a 10% relative lift (4% to 4.4%) at 95% confidence and 80% power, the math asks for 39,475 users per variant, which takes 14 days. Aiming at 15% relative instead (4% to 4.6%) drops the requirement to 17,943 per variant and closes the test in 7 days.

The 14-day experiment involves roughly 78,950 users across both variants. At an instrumentation of 10 events per user (screen view, main clicks, conversion), that is approximately 790,000 events, or 40% of the free monthly allowance of 2 million. Even adding the product’s ordinary traffic on top, a small operation runs this test without leaving the free plan.

Instrumentation scenario	Experiment events	Fits in the 2 million free?	Cost against Pro overage
10 events per user	~790,000	Yes, comfortably	Zero
25 events per user	~1.97 million	Yes, right at the edge	Zero
60 events per user	~4.74 million	No, needs Pro	Zero inside Pro’s 5 million
100 events per user	~7.9 million	No	~145 dollars of Pro overage

Pro’s overage is 0.05 dollars per 1,000 events, that is 50 dollars per additional million. The practical read: what costs money is not running experiments, it is over-instrumenting. Teams that fire an event for every interface interaction blow through the allowance at moderate traffic, while teams that instrument only what feeds a metric run for years inside the free tier.

Per-event billing versus per-visitor billing

The comparison most buyers actually need is not Statsig against another dev-first platform, it is per-event billing against the per-tested-visitor billing that most marketing-led testing tools use. The two models reward opposite behaviors, and knowing which one you are on changes how you plan a program:

	Per event (Statsig)	Per tested visitor (most CRO suites)
What raises the bill	Instrumentation depth, product traffic, analytics usage	How many people you put into experiments
Cost of running more tests	Near zero, since flags are unbilled and the events already exist	Direct, since every tested visitor counts against the plan
Cost of a longer, better-powered test	Near zero on the experiment itself	Direct, the sample size is literally the billing unit
The bad incentive it creates	Under-instrumenting, or arguing about event volume with engineering	Under-powering tests, or testing fewer pages to stay in plan
Who it favors	Teams with many experiments and disciplined instrumentation	Teams with few, large, high-traffic tests

Read that table before comparing sticker prices. A team running twenty small experiments a quarter can find event-based pricing dramatically cheaper, while a team that instruments every UI interaction and runs two tests a year can find it more expensive than a flat plan, on the same traffic. Our A/B testing tools pricing comparison lays the models side by side across vendors.

Reading the result

Say the test ran to 20,000 users per variant and closed with 800 signups on control (4.00%) against 884 on the variation (4.42%).

Control (A)

VisitorsConversions

Variation (B)

VisitorsConversions

Confidence90%95%99%

Control (A) · Rate-

Variation (B) · Rate-

Relative lift-

p-value-

95% CI of the difference-

Two-sided two-proportion z-test. "Not significant" almost always means not enough sample, not that the versions are equal.

Running those four numbers through this blog’s significance engine: z = 2.09, p-value ≈ 0.0365, with a 95% confidence interval on the difference of +0.03 to +0.81 percentage points, and an observed relative improvement of +10.5%. It is significant against the 5% bar, and the interval is wide: the true lift could be almost nothing or nearly double the point estimate. No tool, however good, fixes that; only more sample does, which is why our statistical significance guide insists on reading the interval before the point estimate.

Real strengths

• A free tier you can actually use. Two million events and 50,000 replays per month puts a small operation into production without a contract.
• Public per-event pricing. You can project cost from your own instrumentation instead of going through a quote.
• Flags are not billed. For products with many flags and high traffic, that removes a cost line that exists at some competitors.
• Warehouse-native on Enterprise. For companies with centralized data governance, running the computation on their own warehouse solves a real trust problem about the number.
• Experimentation and analytics on the same events. It avoids the classic divergence between the testing tool’s number and the analytics tool’s number.

Limitations and watch-outs

• Two changes of control in under a year. This is the number one watch-out in 2026, and the reason for short contracts and exportable data.
• Not a tool for marketing without engineering. With no visual editor as the primary path, testing a landing page without a developer is not the product’s natural flow.
• Cost driven by instrumentation. An event architecture decision made by engineering turns into a line on the invoice, and those two teams rarely discuss it beforehand.
• Statistical learning curve. A platform with advanced experimentation features assumes a team that knows what it is switching on; turning everything on without understanding it manufactures false confidence faster than a simpler tool does.
• Converging roadmap. The official announcement talks about integrating the two platforms’ roadmaps. Integration usually brings gains, and also deprecations.

How to evaluate it in a pilot

1. Instrument only what feeds a metric. Do the events-per-user arithmetic before writing the first track call.
2. Run an A/A test. With no real difference between sides, confirm the tool reports the non-effect and that traffic splitting matches what you configured.
3. Test the export path. Get the raw experiment data out before you depend on it.
4. Confirm the current allowances on the official page. Prices and plan limits change, and more so on a platform that has changed owners.
5. Agree an overage ceiling with finance. 0.05 dollars per 1,000 events is cheap per unit and adds up fast with generous instrumentation.

Statsig review: final analysis and who it is for

It fits product and engineering teams that already work with feature flags in their delivery flow, that want experimentation and analytics over the same events, and that benefit from a large free tier to start without a contract. For larger companies with data governance requirements, Enterprise’s warehouse-native execution is a strong argument on its own. If you are weighing it against the open-source side of the category, our GrowthBook vs Statsig comparison covers that decision directly, and our A/B testing tools pricing comparison puts the numbers side by side.

It fits less well for marketing teams that need to test pages without depending on a deploy, for buyers who put long-term contract stability above technical capability, and for small ecommerce operations whose case is web page testing, where client-side tools with a visual editor, like those covered in our Convert.com review and in Google Optimize alternatives, solve it with less effort.

If your case is that second group, Donnu is a lighter option focused on web A/B testing with predictable pricing and conservative statistics. It does not replace Statsig for feature flags, product analytics, session replay, or warehouse execution, and saying otherwise would be dishonest. If what you need is to test pages rigorously without building a platform, start a free 14-day trial.

---

Read also: CRO Tools Compared · Feature Flag Tools Compared · GrowthBook vs Statsig · A/B Testing Tools Pricing Compared

References

• Statsig. Pricing. Developer, Pro, and Enterprise plans, event and replay allowances, overage price, and per-plan features, checked on 14 August 2026. statsig.com/pricing.
• CNBC. OpenAI buys Statsig for $1.1 billion, hires CEO as applications exec. Coverage of the 2 September 2025 acquisition announcement. cnbc.com.
• TechCrunch. OpenAI acquires product testing startup Statsig and shakes up its leadership team. 2 September 2025. techcrunch.com.
• Amplitude. Amplitude and Statsig Partnership. Official announcement of 5 May 2026, with the commitment to maintain and develop the platform and support existing customers. amplitude.com.
• MarTech. Amplitude and Statsig deal raises questions for customers. Independent coverage of the questions the deal raises. martech.org.
• Kohavi, R., Tang, D. and Xu, Y. Trustworthy Online Controlled Experiments: A Practical Guide to A/B Testing. Cambridge University Press, 2020. Chapters on platform choice, A/A testing, and instrumentation. Companion material at experimentguide.com.