AI Models Drive Real Cars: DrivingBench Releases New Benchmark
AI Models Drive Real Cars: DrivingBench Releases New Benchmark
The new DrivingBench benchmark tests AI models' ability to control a real vehicle. GPT-6 Astra was the only model to successfully complete the track after analyzing its errors.
The team behind the DrivingBench project has released a new benchmark in which AI models control a real car.
Models GPT-6 Astra, Claude Fable 5.1, and Grok 4.6 were connected to a Toyota Corolla, given control over the steering wheel, gas, and brake pedals. Each model was allowed up to three attempts to complete the track.
Only GPT-6 Astra managed to complete the course successfully. In the first attempt, the model covered 49% of the route. After being prompted to analyze its mistakes, Astra completed the entire track.
For more details on the test methodology and results, visit the project website: DrivingBench.
Why it matters
AnalysisThis test highlights the progress in multimodal AI models' ability to make real-time decisions and self-correct, which is a significant step forward for autonomous driving technology.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
OpenAI introduces GPT-6 Cyber, a record-setting model for cybersecurity tasks
OpenAI reports new cybersecurity benchmark results for GPT-6 Cyber, including a 48.5% score in ExploitGym and the discovery of 81 previously unknown Firefox vulnerabilities, while access remains limited.
Thinking Machines Releases Inkling 2 Small: a 594B MoE Model That Can Fine-Tune Other Models
Thinking Machines released Inkling 2 Small, a 594B-parameter MoE model that leads among open models in several benchmarks and demonstrated autonomous fine-tuning of Qwen3.5-4B.
Frontier Models
The most powerful class of artificial intelligence at the forefront of global research (Claude 3.7 Sonnet, OpenAI o3/GPT-4.5, Gemini 2.0 Pro), defining the limits of modern reasoning, autonomy, and coding capabilities.
Part I. AI Agent Architecture: The Foundations
The eleven foundational building blocks of modern autonomous agents: model taxonomy, the agentic runtime loop, advanced prompting, distillation, skills architecture, memory systems, function calling, and FinOps.