Torki 2 Benchmarks
Torki 2 on 22 benchmarks, compared with leading models
AI assistants, AI agents, coding agents, voice AI, multimodal AI, AI research, evaluation and benchmarks
TorkeHub makes Torki, one AI assistant that answers, researches, makes images, builds prototypes and decks, runs code and takes on work you hand it — and Torki Code, a coding agent for real software work.
Torki runs on our own models, on machines we own and operate, and it works on the web, iPhone, iPad, Android and Mac, all on one account.
| What it does | |
|---|---|
| 💬 Chat | Ask anything, with web search and four modes — from Pulse (fast) to Cosmos (maximum capability). |
| 🔎 Research | Searches first, then answers with its sources. |
| 🎙️ Voice | Talk to Torki in real time, with seven companions to choose from. |
| 🖼️ Vision | Reads photos, screenshots and charts, and makes images. |
| 🧩 Forge | Turns a description into a landing page, dashboard, prototype or pitch deck — no code needed. |
| 🧑💻 Canvas | Writes and runs code and documents side by side. |
| 🤖 Agents | Hand off multi-step work and let it run. |
| 🗂️ Projects | Keep files, chats and instructions together. |
Ask in plain language and Torki Code plans the change, edits the files, runs the commands and tests, and reports back. On Mac today, with Windows and Linux coming.
Torki 2's benchmark results are here: 22 public benchmarks, measured on the same model and settings Torki users get, with exactly how each one was run and the per-question result behind every score.
Follow TorkeHub to see them first.