A researcher needs to compare how four different language models interpret the same dataset. She's opening browser tabs, copying prompts, pasting into separate chat windows, then manually comparing outputs in a spreadsheet. She's burning hours on context-switching between interfaces. 120.dev runs multiple AI conversations side-by-side in one window, letting her feed identical prompts to different models and see responses in parallel. The interface hits 120 frames per second, built natively for Mac and Windows rather than as a web app wrapped in a shell.
This application connects to over 500 AI models through a single interface. A developer testing prompt variations can query OpenAI's latest models, Anthropic's reasoning engines, Google's multimodal systems, and xAI's offerings without juggling accounts. Everything runs through your own API keys. No subscription fees. You pay only what the API providers charge directly. All conversation history saves locally on your device. Nothing goes to 120.dev's servers. Zero data collection.
Local model support matters for privacy-focused teams. A legal analyst working with confidential documents can run Llama or DeepSeek through Ollama or LM Studio, keeping everything on their machine. They upload PDFs in batches, chat with the content, and the files never leave their computer. This works until they need web search or real-time internet access. Those features require cloud models.
The thinking mode adjusts reasoning depth across three levels. Low runs fast for simple queries. High takes longer but works through complex logic step-by-step. A financial analyst building valuation models switches between depths depending on whether she's checking basic calculations or stress-testing assumptions across scenarios.
Image generation pulls from multiple sources. A product designer generates concepts with Stable Diffusion, then edits variations with other models, comparing outputs without switching applications. Text-to-speech reads responses aloud. The prompt library stores templates. Code renders with syntax highlighting across different themes.
It breaks for teams. No collaboration features. No shared workspaces. One license covers one person. A content studio can't split the cost across writers or maintain shared prompt libraries. The free plan restricts commercial use and limits you to two basic models for 30 days. After that, you need the Pro license at $78 one-time to access the full model roster, document chat, web search, and advanced generation tools. That license includes updates for one year.
The Linux version doesn't exist yet. Mac and Windows only. A developer running Ubuntu on their primary machine can't use it without spinning up a VM or dual-booting.
Temporary conversations don't save history. A consultant testing sensitive client scenarios can chat without creating records. Branching lets you fork conversations at any point, exploring different prompt paths from the same starting context. Split chats run multiple threads from one input.
This works best for individuals who want model flexibility without monthly fees. Researchers comparing outputs with developers testing across providers. Privacy-conscious professionals keeping data local. It doesn't fit teams needing shared access or users locked into Linux. The bring-your-own-key model means unpredictable API costs. You control spending but shoulder the complexity of managing multiple provider accounts.
