Benchmarks

AI geolocation, measured.

A closer look at the accuracy, speed, and cost behind our homepage charts. One shared image dataset, with the mode and pricing basis made clear for every result.

Download datasetExplore methodology

The results at a glance

Worldwide image geolocation · 62-image custom dataset

Swipe the table to compare all metrics →

October 14, 2026 benchmark results. Accuracy uses each provider’s most accurate publicly available worldwide image-search mode or model; latency uses its fastest. Costs are rounded averages.
ProviderAccuracy @ 1 km Most accurate mode/model · higher is betterLatency · p50 Fastest mode/model · lower is betterCost / search USD · lower is better
GeoSeer94%Agent10 sFast$0.20Rounded average
GeoAxis87%Global Search · High35 sGlobal Search · Low$0.90Rounded average
Oceanir79%D3 · Forensic8 sD2 · Standard$1.30Rounded average
Picarta62%Standard image search12 sStandard image search$0.70Rounded average

Each column answers a separate question. The accuracy mode, fastest mode, and average search cost describe different operating points; they do not represent one combined search configuration.

Why these providers?

We selected Picarta, Oceanir, and GeoAxis as three prominent AI geolocation services, based on their public visibility and geolocation capabilities at the time of testing. The comparison focuses on self-serve, worldwide image searches. GeoAxis’s city-restricted search is outside this scope because it requires a known city.

We exclude private or enterprise-only solutions, including GeoSpy (now Raven) and Palantir’s enterprise platforms, which fall outside this public, self-serve comparison. Enterprise-only models within the listed providers are also excluded.

This is GeoSeer’s own evaluation, rather than an independent market-wide ranking.

The methodology

Three metrics. Clear definitions.

All accuracy comparisons use the same custom dataset linked on our homepage: 62 images adapted and expanded from Im2GPS, spanning urban, rural, and indoor scenes with verified reference coordinates. Download the image set to inspect the inputs.

01 / Accuracy

Each provider’s most accurate mode/model

Accuracy @ 1 km is the share of test images whose top predicted coordinate falls within 1 kilometre of the verified location, measured as geographic distance. A country or city match alone does not meet this threshold.

For accuracy, we select each provider’s most accurate publicly available mode or model for worldwide image search: GeoSeer Agent, GeoAxis Global Search at High effort, Oceanir D3 Forensic, and Picarta’s standard image search. Speed is not the selection criterion for this metric.

Accuracy @ 1 km = correct predictions within 1 km ÷ test images × 100

02 / Latency

Each provider’s fastest mode/model

p50 latency is the median time from search submission until a completed location result is available. Half of completed searches finish at or below this time. Values are reported in seconds.

For latency, we select each provider’s fastest publicly available mode or model for worldwide image search: Oceanir D2 Standard (8 s), GeoSeer Fast (10 s), Picarta’s standard image search (12 s), and GeoAxis Global Search at Low effort (35 s). These selections are made separately from the accuracy test. City-restricted retrieval is excluded to keep the worldwide scope consistent.

These are the October 14 comparison results. Provider-published timings in the references may use different inputs or conditions. Network conditions, queues, and scene complexity can change individual response times.

03 / Cost

Average paid cost for one image search

Costs are rounded USD estimates per search. We use the paid credit or plan references below, averaging modes equally where their costs differ; Picarta uses the midpoint of its two listed paid search rates. Plan-based estimates assume full utilization. Free allowances, promotions, and negotiated enterprise rates are excluded.

  • GeoSeer · $0.20: One-time credits: $1 for 5 searches ($0.20 each), available in the Dashboard. Fast and Agent image searches each use one credit, so their average is $0.20. Subscription pricing is not used.
  • GeoAxis · $0.90: Pay-as-you-go at the published $0.90 per search; the minimum $10 purchase covers about 11 searches. This is the reference rate across Global Search effort levels; subscription pricing is not used.
  • Oceanir · $1.30: Pro at $29/month for 100 credits ($0.29 per credit). D1, D2, and D3 use 1, 2, and 10 credits: $0.29, $0.58, and $2.90 per run. Their equal-weight average is about $1.26, rounded to $1.30, assuming full credit utilization.
  • Picarta · $0.70: Monthly subscription at $59.90 for 100 searches (about $0.60 each at full utilization) and a one-time wallet at $15.90 for 20 searches (about $0.80 each). The midpoint of these two paid rates is about $0.70; annual billing is not used.

Oceanir’s one-time top-up packs ($7.99 for 10 credits to $59.99 for 100) are not used in this calculation. Prices were checked against the GeoAxis, Oceanir, and Picarta pricing pages on October 16, 2026; benchmark measurements remain the October 14 snapshot.

These figures compare consumer search economics. Commercial API access and enterprise usage can require separate plans or agreements.

How to read this snapshot

GeoSeer has the highest 1 km accuracy and lowest average cost in this comparison; Oceanir has the fastest median response. Results describe this small, custom image set and the configurations listed above. They do not establish performance on every geography, scene type, video, or event search. Percentages are rounded summary results; per-image run logs and statistical uncertainty intervals are not included in this report.

Product & pricing references

Official documentation for modes and pricing. The comparison results above are GeoSeer’s October 14 snapshot.

GeoSeer pricing GeoAxis search modes GeoAxis pricing & paid searches Picarta pricing Oceanir analysis depths Oceanir plans & credits Oceanir timing reference

Put the results to work.

Start with your own image and explore the location evidence.

Try GeoSeer