The Navigator API defaults to Navigator n2, Yutori’s flagship computer-use model for full desktop environments. It works across GUIs, browsers, terminals, tools, and code, with frontier-level accuracy across OSWorld 2.0, OSWorld-Verified, WeaveBench, MyPCBench, and MacAgentBench. Navigator n1.5 is Yutori’s computer-use model for browser automation. It drives a browser from screenshots and action history. It scores an independently verified 97.3% (human evaluation) on Online-Mind2Web and 89.7% averaged across Navi-Bench v2 and Westworld, outperforming frontier models from OpenAI, Anthropic, Google, and Meta on web-task success — while being 2–5x cheaper.

Navigator n2 delivers frontier-level accuracy across a diverse set of computer-use benchmarks while costing $0.50 per 1M input tokens, $0.05 per 1M cached input tokens, and $4.00 per 1M output tokens. On web-task benchmarks, Navigator n1.5 outperforms Anthropic’s Claude, OpenAI’s GPT, Google’s Gemini, and Meta’s Muse at roughly 2.93 seconds per step and $0.81 per task, delivering higher accuracy at lower latency and cost.

Yes. Navigator follows the OpenAI Chat Completions interface, so you can call it with the OpenAI Python or JavaScript SDK by pointing the base URL at api.yutori.com. Use the n2 model for general computer use across full desktop environments, or Navigator n1.5 for browser-optimized workflows. Yutori also provides a Python SDK and a hosted MCP server.

Yes. Pass a JSON schema and Navigator n1.5 returns parsed JSON that conforms to it, and it can execute arbitrary JavaScript and read the DOM for reliable extraction from dynamic pages. See the Navigator n1.5 reference. For full-desktop workflows, Navigator n2 can combine mouse and keyboard actions with shell commands and short snippets of code, choosing the right interface as the task progresses.

Navigator n2 costs $0.50 per 1M input tokens, $0.05 per 1M cached input tokens, and $4.00 per 1M output tokens. Navigator n1.5 costs $1.50 per 1M input tokens and $5.00 per 1M output tokens. New accounts receive $5 in free credits. On web-task benchmarks, it delivers higher accuracy at lower latency and cost than frontier computer-use models. See the full pricing.