# Mapbox says Traffic 2.0 gets arrival times right on 98% of trips By MarketScale Newsroom · Published 2026-09-28 · Industrial IoT on MarketScale Canonical: https://www.marketscale.com/industries/industrial-iot/mapbox-says-traffic-20-gets-arrival-times-right-on-98-of-trips > At BUILD, Mapbox introduced Traffic 2.0, a Places API in public preview, and agent tools. The 98% ETA claim is what fleet teams can test first. ## Key points - Traffic 2.0 claims accurate ETAs on 98% of trips, a vendor figure fleet teams can benchmark against ETA accuracy on their own lanes. - An AI agent that reroutes a driver is only as useful as the traffic data underneath it; Mapbox's agentic mapping engine processes anonymized movement data from over 45,000 applications. - The Places API, now in public preview, combines business listings with entrances, building footprints, operating hours and visitation patterns; leaving preview is one signal to watch. Mapbox says its new Traffic 2.0 engine gives an accurate estimated time of arrival on 98% of trips. The launch is built around AI agents, but that one number is what a dispatch or fleet team can actually hold the company to. Mapbox introduced the new services on Sept. 17, according to its announcement distributed through PR Newswire. CEO Peter Sirota unveiled the capabilities in his opening keynote at Mapbox BUILD. Mapbox's argument is simple: AI models can talk about almost anything, but they have a weak grasp of physical location, so their recommendations come out generic and ignore what is actually around the user. ## How Traffic 2.0 checks its own work The engine's models learn by comparing the arrival times they predicted with the times vehicles actually arrived. That feedback loop runs inside what Mapbox calls its agentic mapping engine, which combines live movement with historical patterns. Traffic 2.0 also accounts for congestion at major maneuvers along a route. A model that corrects itself this way suggests accuracy should track the volume of real trips flowing back into it. An AI agent that reroutes a driver is only as useful as the traffic data underneath it. Mapbox's agentic mapping engine relies on autonomous agents that work through live inputs plus anonymized movement data drawn from more than 45,000 applications. That processing is how Mapbox keeps its roads, traffic conditions and map data up to date in close to real time. Teams that already measure ETA accuracy on their own lanes now have a clean vendor figure to test against. ## Places, down to the entrance The second pillar is the Mapbox Places API, now in public preview. It combines standard business listings with the physical details a driver or technician needs on arrival: building footprints, operating hours, entrances and visitation patterns. Block Field Entrances and footprints are what separate arriving at a building from arriving at the right door. For companies whose apps send people to commercial sites, one API carrying hours and entrances alongside the address could replace several separate lookups. Places is still in public preview, so for now it's software to evaluate. ## Agents that can edit a route The Agent Toolkit for Mapbox's Maps and Navigation SDK opens those controls to AI agents. Mapbox's examples are specific: a driver uses voice commands to add a stop or check how traffic will affect an upcoming meeting, and a chat agent drops points on a map, moves the view or summarizes a route. The Notion integration could be the piece that lands closest to an operations desk. Notion Agents can now call Mapbox tools inside workspace documents to clean and standardize customer addresses, track deliveries and coordinate field operations. Address cleanup is dull work, which is exactly why automating it where teams already keep their customer lists is appealing. Two more features face the end user. Natural Language Queries Search extends the Search Box API to handle conversational requests, and Mapbox's example is nearby coffee shops with fast Wi-Fi and a quiet place to work. AI Mode for the Static Images API restyles a map image from a text prompt when it's requested, a snowy Paris for instance, while keeping the real roads and buildings of the basemap. Block Field ## Cheaper to try, easier to wire in The new Mapbox connector for Figma puts real maps directly into Figma designs, so screenshots and developer handoffs are no longer needed. In Figma's chat, a designer describes the map they want, for example "a map of central Copenhagen, dark theme." The connector then generates the map with the location, zoom level, style and time of day the designer asked for, and it can add routes, pins and areas on top. ## Where the pitch meets the dispatch floor Two signals will show whether that core holds up. The first is the Places API leaving public preview. The second is customer or independent measurement of Traffic 2.0 against the 98% claim on real delivery routes, the only test that tells a fleet manager whether the new ETAs change the arrival time a customer is told to expect. ## Sources - [Mapbox introduces location infrastructure for AI](https://www.gpsworld.com/mapbox-introduces-location-infrastructure-for-ai/) (GPS World) - [Mapbox Announces Location Infrastructure for AI](https://www.mapbox.com/press-releases/mapbox-announces-location-infrastructure-for-ai) - [Mapbox Announces Location Infrastructure for AI](https://www.prnewswire.com/news-releases/mapbox-announces-location-infrastructure-for-ai-mapbox-announces-location-infrastructure-for-ai-302882326.html) (PR Newswire) - [Location AI Overview | Help - Mapbox Documentation](https://docs.mapbox.com/help/getting-started/location-ai/) - [How Mapbox is Quietly Powering the AI Map Revolution with Kieran ...](https://www.youtube.com/watch?v=61awIIToFO0) Tags: Mapbox, Traffic 2.0, Mapbox Places API, AI agents, location data, logistics, fleet management, last-mile delivery, navigation, developer tools, Notion, IT operations --- Source: MarketScale, https://www.marketscale.com/industries/industrial-iot/mapbox-says-traffic-20-gets-arrival-times-right-on-98-of-trips. Published for AI indexing and citation; cite the canonical URL. Site guide for agents: https://www.marketscale.com/llms.txt