Alibaba's Amap Updates ABot: Five Robotics Models and SOTA on 17 Benchmarks
Amap, Alibaba's mapping service, has announced a full overhaul of its ABot embodied-AI system and released five models at once: ABot-N1, ABot-M0.5, ABot-ER, ABot-AgentOS and ABot-C0. The update covers a robot's perception, decision-making, execution and memory. According to the company, ABot reached SOTA on 17 leading benchmarks and has become more accurate at carrying out general-purpose tasks autonomously.
AI-processed from 36Kr (36氪); edited by Hamidun News
Amap (Gaode, 高德), the mapping service of Alibaba Group, announced a full overhaul of its ABot embodied-AI system in July 2026 and simultaneously released five new models that took first place (SOTA) on 17 leading benchmarks.
Which models did Amap release?
Amap unveiled five embodied-intelligence models under the ABot brand. It is the first time Alibaba's mapping service has rolled out a whole lineup of models for controlling robots. The company calls the update "full-stack": it touches all key layers of the system rather than a single module.
- ABot-N1, ABot-M0.5, ABot-ER, ABot-AgentOS, ABot-C0 — five models presented at the same time
- SOTA (state-of-the-art) on 17 leading benchmarks
- The update spans four domains: perception, decision-making, execution and memory
- Developer: Amap (Gaode), a unit of Alibaba Group
- Stated goal: autonomous execution of general-purpose tasks by robots
What does the update cover?
The ABot update touches four domains of robot operation: perception, decision-making, execution and memory. By this logic, the five-model lineup covers the full loop, from how a robot "sees" its surroundings to how it remembers previous actions and plans the next ones.
Amap has not disclosed the specific role of each of the five models or their individual technical parameters. In its statement, the company stresses the systemic approach: by updating all layers at once, ABot is meant to improve, across the board, robots' ability to perform general tasks on their own.
"A unified update spanning perception, decision-making, execution and memory comprehensively improves robots' ability to autonomously perform general-purpose tasks,"
Amap said in its statement.
Why does it matter?
Amap's move into embodied AI shows that Alibaba is extending its robotics bet beyond the cloud and language models. Amap is known primarily for navigation and maps, so a lineup of five models for physical robots marks a notable turn toward "embodied" intelligence that controls machines in the real world, not just text on a screen.
The result on 17 benchmarks is a signal to competitors: Chinese tech giants are building up their presence in embodied AI, where labs are racing to control humanoids and service robots. What matters for the industry is that a major internet player is releasing five models at once and claiming leadership on independent metrics.
What does this mean?
Amap has staked out a position in embodied AI with five models and SOTA results on 17 benchmarks, a bid by Alibaba for a role in the race for autonomous robots. How strong that bid is will be shown by independent verification and real-world deployments, details of which the company has not yet disclosed.
Frequently asked questions
What is ABot from Amap?
ABot is an embodied-intelligence system from Amap, a unit of Alibaba, for autonomous robot control. It received a full overhaul in July 2026 and includes five models: ABot-N1, ABot-M0.5, ABot-ER, ABot-AgentOS and ABot-C0.
On how many benchmarks did ABot reach SOTA?
According to Amap, the updated ABot system reached SOTA (state-of-the-art) on 17 leading benchmarks covering a robot's perception, decision-making, execution and memory.
- Oct 28, 2026Wired Tested 6 AI Agents in a Day: Dots Leads, Gemini Spark Fails at Lunch
- Oct 28, 2026Anthropic reveals how it builds Claude with Claude: the brain separate from the hands
- Oct 28, 2026AI Agents Formalize 15 000 Pages of Proofs in 24 Days and Find a Gap
- Oct 28, 2026Anthropic Turns Claude Code Into a Platform: Claude Mods Flip the Agent Harness
Key Questions Answered
Which models did Amap release?
Amap unveiled five embodied-intelligence models under the ABot brand. It is the first time Alibaba's mapping service has rolled out a whole lineup of models for controlling robots. The company calls the update "full-stack": it touches all key layers of the system rather than a single module. ABot-N1, ABot-M0.5, ABot-ER, ABot-AgentOS, ABot-C0 — five models presented at the same time SOTA (state-of-the-art) on 17 leading benchmarks The update spans four domains: perception, decision-making, execution and memory Developer: Amap (Gaode), a unit of Alibaba Group Stated goal: aut
What does the update cover?
The ABot update touches four domains of robot operation: perception, decision-making, execution and memory. By this logic, the five-model lineup covers the full loop, from how a robot "sees" its surroundings to how it remembers previous actions and plans the next ones. Amap has not disclosed the specific role of each of the five models or their individual technical parameters. In its statement, the company stresses the systemic approach: by updating all layers at once, ABot is meant to improve, across the board, robots' ability to perform general tasks on their own. "A unif
Why does it matter?
Amap's move into embodied AI shows that Alibaba is extending its robotics bet beyond the cloud and language models. Amap is known primarily for navigation and maps, so a lineup of five models for physical robots marks a notable turn toward "embodied" intelligence that controls machines in the real world, not just text on a screen. The result on 17 benchmarks is a signal to competitors: Chinese tech giants are building up their presence in embodied AI, where labs are racing to control humanoids and service robots. What matters for the industry is that a major internet player
What does this mean?
Amap has staked out a position in embodied AI with five models and SOTA results on 17 benchmarks, a bid by Alibaba for a role in the race for autonomous robots. How strong that bid is will be shown by independent verification and real-world deployments, details of which the company has not yet disclosed.
Delegate Deep Research, Coding & Multi-Step Workflows to Nomad Agent
Deploy autonomous task execution with Nomad Agent (agent.nomadnet.ai) — automated web research, data synthesis, code generation, and MCP tool orchestration.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.