Mapbox vs Google vs HERE Geocoding

This article extends the Geocoding sub-hub with a direct head-to-head comparison of the three dominant commercial geocoders. Companion to /learn/major-geocoding-services-compared (the broader survey) and /learn/geocoding-accuracy-levels.

The three at a glance

| Aspect | Google | Mapbox v6 | HERE | | ------ | ------ | --------- | ---- | | Pricing (above free tier) | ~$5/1,000 | ~$0.50/1,000 | ~$0.50-1/1,000 | | Free tier | $200/month credit (~40k req) | 100,000/month | 250,000/month | | Caching policy | 30-day allowed | Not retained (per ToS) | 30-day typical | | Global accuracy | Best | Strong (NA, Europe) | Strong (Europe especially) | | POI database | Best | Strong | Strong (some regions) | | Autocomplete UX | Strong | Strong (v6 designed for it) | Solid | | Batch endpoint | No | Yes (v6) | Yes | | Confidence bands | Limited | Yes (low/medium/high) | matchLevel + score | | Map dependency | Required for some uses | Independent | Independent | | Logistics integration | Limited | Limited | Strong (Navteq lineage) |

For Coordinately's use case (web app, modest volume, user-typed queries), Mapbox v6 is the operational sweet spot. The Coordinately tools use Mapbox v6 for forward/reverse geocoding (see arch decisions §19.1).

Google Maps Geocoding

The most accurate, most expensive, and most ubiquitous geocoder.

Pricing

Google operates on a monthly $200 credit model that translates to:

• First ~40,000 geocoding requests/month: free (covered by the credit).
• Above 40,000: $5 per 1,000 requests.

For high-volume use cases, costs grow quickly: 1 million geocodes/month is ~$5,000/month.

Accuracy

Google's geocoder is typically the most accurate globally. Especially strong in:

• North America: comprehensive street-level coverage including new construction.
• Western Europe: excellent address and POI coverage.
• Major Asian markets: Japan, South Korea, Singapore — Google leads.
• POI database: Google's business listings database (the largest commercially) gives unmatched POI coverage.

In some markets (sub-Saharan Africa, parts of Central Asia), Google's coverage advantage is less pronounced.

Caching policy

Google's Terms of Service allow caching for up to 30 days:

• Store lat, lng, place_id for up to 30 days.
• Refresh by re-querying before the period expires.
• This is more permissive than Mapbox (no retention) and similar to HERE (30-day typical).

For applications with repeating queries (same addresses, common locations), the caching policy enables significant cost reduction.

Restrictions

Google's Terms include several restrictions:

• Display constraint: results must be displayed on a Google Map (with exceptions for specific use cases — see Google's detailed terms).
• No bulk geocoding: large pre-geocoding projects need Google's commercial Maps Platform agreement.
• No display in competing applications (depending on the API and license).

When Google is the right choice

• Best-in-class global accuracy is the priority.
• POI matching is critical.
• Already integrated with Google Maps Platform.
• Budget for premium pricing.
• 30-day caching materially reduces cost (e.g., e-commerce with repeating customer addresses).

Mapbox Geocoding v6

The web-app sweet spot. Released in 2023 as a substantial redesign of Mapbox's geocoder.

Pricing

Mapbox uses simple tiered pricing:

• First 100,000 requests/month: free.
• Above 100,000: $0.50 per 1,000 requests (for v6 geocoding).

For a moderate web app doing 50,000 geocodes/month: free. For 1 million/month: ~$450/month — about 1/10 of Google's cost at that volume.

Caching policy: no retention

Mapbox's Terms of Service explicitly state that geocoding response data must not be retained:

“...you may not store or cache geocoding response data...”

This affects application architecture: every query must hit the API; no local caching of common results. Per Coordinately's architectural decisions (§19.2), this is compatible with per-request SSR (server renders fresh on each user request) but forbids persistent caching or pre-baking address tables.

For applications with mostly unique queries (typical web app type-ahead, user-entered addresses), the no-retention rule isn't a meaningful constraint. For applications with repeated queries (same addresses geocoded many times), the lack of caching is a significant cost driver.

Features

Mapbox v6 includes:

• Confidence bands (high, medium, low) on every result. Coordinately's arch decision §19.1 specifies that low/medium confidence results should be surfaced as candidates, not auto-selected.
• Structured input: separate query fields for street, city, state, postcode rather than free-form.
• Autocomplete optimization: designed for type-ahead suggestion UX.
• Batch endpoint: process up to 1,000 addresses per request.
• Country biasing: prefer results from specific countries.
• Language preferences: results in user's preferred language.
• Types filter: limit results to place / address / POI / region.

Accuracy

Mapbox v6 is strong in North America and Western Europe — comparable to Google for street-address geocoding. Slightly weaker for POI matching in some regions (Google's business-listings database advantage). In less-mapped regions (sub-Saharan Africa, some Central Asian states), coverage is more limited than Google.

For typical web-app geocoding use cases (find an address the user types), Mapbox v6 is typically indistinguishable from Google in accuracy.

When Mapbox is the right choice

• Web application with moderate geocoding volume.
• Cost-conscious (1/10 of Google at scale).
• No-retention compatible with the application architecture (per-request SSR is fine; batch preprocessing of addresses is forbidden).
• Modern API design (confidence bands, autocomplete, structured input).
• Already using Mapbox or MapLibre for the base map (single vendor relationship).

HERE Geocoding

The European-strength alternative. Built on the Navteq → HERE lineage of traditional GPS map data.

Pricing

HERE's pricing varies by plan tier:

• Freemium: 250,000 transactions/month free.
• Pay-as-you-go: $0.50-1 per 1,000 above the free tier.
• Enterprise: negotiated pricing for high volume.

For a 100k/month application: free. For 1M/month: ~$375-750/month — comparable to or slightly better than Mapbox at high volume.

Accuracy

HERE is strongest in Europe — the Navteq inheritance shows in detailed German, French, UK, Italian, Scandinavian coverage. Address-level accuracy in European cities is often the best of any commercial geocoder.

In North America, HERE is comparable to Mapbox and slightly behind Google. In less-mapped regions, coverage varies but generally lags Google slightly.

Features

• Batch endpoints: explicit batch geocoding API for high-volume processing.
• Routing integration: tight integration with HERE's routing services (for logistics use cases).
• Reverse geocoding with rich address details.
• Place search: dedicated place/POI search separate from address geocoding.
• Autocomplete API: separate autocomplete API optimized for type-ahead UX.
• matchLevel reporting: per-result match quality metadata (similar to Mapbox confidence bands but with more granularity).

Caching policy

HERE generally allows 30-day result caching, similar to Google. Specific terms depend on the license plan; enterprise plans may have different provisions.

When HERE is the right choice

• European user base (strongest accuracy).
• Logistics application (routing integration, POI-to-address matching).
• Batch geocoding needed (explicit batch API).
• Moderate to high volume, cost-sensitive (generous free tier, competitive paid pricing).
• Caching benefits the cost model.

Specific scenarios

Scenario 1: Web app with user-typed addresses

Moderate volume, mostly unique queries. Mapbox v6 is typically optimal: strong UX (autocomplete, confidence bands), low cost (likely free at modest volume), no-retention rule isn't a constraint.

Scenario 2: E-commerce with repeating addresses

Many users in similar locations. Google or HERE with 30-day caching can dramatically reduce costs. Coordinately's no-retention Mapbox doesn't benefit here.

Scenario 3: Logistics / last-mile delivery

POI-to-address matching matters; batch processing needed. HERE is the typical choice for the strongest fit.

Scenario 4: Best-in-class accuracy regardless of cost

Google. Premium pricing pays for the global accuracy and POI database depth.

Scenario 5: Open-source / cost-zero requirement

Nominatim (self-hosted). See /learn/nominatim-explained.

Scenario 6: US address validation

SmartyStreets. Specialty US-only service with deeper US address parsing than any of the global geocoders.

Scenario 7: Enterprise GIS / government

Esri ArcGIS World Geocoding. Premium service integrated with the ArcGIS ecosystem. Used extensively by US government agencies.

Other major options

Beyond the three primary providers:

Esri ArcGIS World Geocoding

Premium service integrated with ArcGIS. Strong in US, especially government use. Enterprise pricing.

Amazon Location Service

AWS's managed geocoding service, with HERE and Esri as backends. Integrated billing through AWS account.

TomTom

Strong in European logistics. Recently expanded consumer-API offering. Competitive pricing.

Geoapify, LocationIQ

Hosted Nominatim with commercial SLAs (see /learn/nominatim-explained).

SmartyStreets

US-only specialty service. Best-in-class for US address validation, parsing, ZIP+4 matching, USPS verification.

OpenCage

Aggregator service that combines multiple geocoders (Mapbox, Nominatim, others) into a single API. Useful for redundancy or cross-checking results.

Comparison axes

When evaluating geocoders for a specific use case, the relevant axes:

Accuracy by region

Test the geocoders against your actual address samples. Don't rely on general comparisons — specific regions can favor one geocoder unexpectedly. The Mapbox v6 confidence band can be used as a self-reported accuracy indicator (though calibrate against truth before trusting).

Pricing at expected volume

Forecast 12-month volume; compute monthly cost at each tier. Include free tiers; consider growth.

Caching benefits

Are queries repeating (e-commerce, registered users) or unique (web app autocomplete)? Caching changes the cost-benefit dramatically.

API design

Autocomplete optimization, structured input, confidence bands, batch endpoints, language preferences, country biasing. Each provider differs.

Operational integration

Existing vendor relationships, SLAs, monitoring, billing systems. A single-provider stack is simpler operationally.

Compliance and data sovereignty

Some jurisdictions require data residency (EU GDPR, some government clouds). Verify the geocoder's data handling and regional API endpoints.

Multi-provider strategies

Some applications use multiple geocoders:

Primary + fallback

Default to the primary (e.g., Mapbox); fall back to a secondary (e.g., Google) on low-confidence results or specific regions where the primary is weak.

Cross-verification

Query multiple geocoders; flag disagreement for manual review. Useful for high-stakes geocoding (land records, legal addresses).

Regional best-of-breed

Use HERE for European queries, Google for Asian, Mapbox for North American — exploiting each geocoder's regional strengths.

Cost optimization

Use the cheaper provider for the bulk; escalate to the premium provider for low-confidence cases that need verification.

The complexity is meaningful (multiple integrations, billing, monitoring) and usually only worth it for high-stakes or high-volume applications.

Coordinately's choice

Per CLAUDE.md §3 and arch decisions §19.1, Coordinately uses Mapbox Geocoding API v6. Rationale:

• Cost-effective at expected volume (well under the 100k free tier for the foreseeable future).
• Modern API design (confidence bands match the arch §19.1 requirement for surfacing candidates).
• Strong autocomplete for tool UX.
• No-retention rule compatible with per-request SSR architecture (§19.2).
• Single-vendor stack with MapTiler (base map tiles).

The trade-offs: less coverage in some non-Western regions vs Google; not the strongest in Europe vs HERE. But for the Coordinately use case (coordinate-tool web app with global but not deep non-Western emphasis), Mapbox v6 is the right choice.

Common misconceptions

“Google is the most accurate everywhere.” Most regions yes, not everywhere. HERE often wins in Europe; specialty services beat all three in specific contexts (US addresses, military codes).

“Mapbox is just the cheap option.” Mapbox v6 is a serious competitor, not just a budget alternative. The confidence bands, modern API, and v6 redesign make it a first-tier choice for many web-app use cases.

“HERE is only for enterprises.” No — the 250k/month free tier is generous and accessible to small developers. HERE's enterprise positioning is genuine but doesn't exclude smaller users.

“Caching reduces cost equally for all providers.” Provider-dependent: Mapbox no-retention rule eliminates caching as a strategy; Google and HERE permit 30-day caching. For repeat-query applications, this is a significant differentiation.

“Confidence bands are universal.” Provider-specific. Mapbox v6 has explicit high/medium/low; Google has partial_match boolean and location_type enum; HERE has matchLevel + matchQuality. The values are not directly comparable across providers.

“You should never use Google for geocoding.” It depends. Google is expensive but often the most accurate; for specific use cases (POI matching, premium accuracy requirements, existing Google Maps Platform integration), it's the right choice.

“The free tier is enough for production.” Depends on volume. For a moderate web app (under 50-100k geocodes/month): yes. For high-volume production: no — plan for paid usage.

“Geocoder accuracy is a binary thing.” No — it's a spectrum. Different providers have different accuracy profiles by region, address type, POI category, and other dimensions. Test against your specific data.

“APIs are stable.” Pricing, features, and terms change. Mapbox redesigned its API from v5 to v6; Google introduced and modified the $200 credit model; HERE rebranded several times. Long-term applications should plan for API evolution.

“You can use the same geocoder for autocomplete and geocoding.” Yes but optimize. Autocomplete (type-ahead suggestions) is a different UX pattern than full geocoding; many providers have dedicated autocomplete endpoints that are more efficient. Mapbox v6 and HERE have explicit autocomplete APIs distinct from geocoding.

“Geocoders never change their results.” They evolve continuously — new addresses appear, old ones disappear, OSM data refines, commercial data updates. The same query today and a year from now may produce different results.

“Open-source geocoders can't compete with commercial.” In well-mapped regions, Nominatim is competitive. For specialty needs or sparse regions, commercial options have an edge. For most use cases the trade-off is cost (open-source: hosting + ops; commercial: API fees) rather than quality.