Skip to main content

Model tiers

c0mpute offers three tiers, each running a different model on different infrastructure.

Free — Qwen 1.5B​

  • Runs in browser workers via WebGPU
  • ~900MB model download
  • Fast responses, low VRAM requirements
  • Standard model with default safety filters
  • Good for: simple questions, quick lookups, casual use

This is the default tier. It works on most modern devices with WebGPU support.

Pro — Dolphin Mistral 7B​

  • Runs in browser workers via WebGPU
  • ~4GB VRAM required
  • Uncensored — won't refuse topics based on corporate content policies
  • Higher quality reasoning and longer, more detailed responses
  • Workers earn 2x compared to Free tier jobs

Pro uses the Dolphin fine-tune of Mistral 7B, which has been trained without artificial refusal behavior. It answers what you ask without moralizing or deflecting.

Max — Qwen2.5 14B abliterated​

  • Runs on native workers via node-llama-cpp
  • ~9GB VRAM required on the worker's GPU
  • Uncensored — abliterated (refusal behavior surgically removed)
  • Web search — can search the internet and cite sources
  • Best quality responses across all tiers
  • Workers earn 3-5x compared to Free tier jobs

Max is the premium tier. It runs on dedicated native workers with powerful GPUs, delivering the highest quality responses in the network.

What "uncensored" means​

Corporate AI models (ChatGPT, Claude, Gemini) are trained to refuse certain topics. Ask about anything the company considers sensitive and you get a refusal. These aren't safety features — they're content policies imposed by corporations.

Uncensored models like Dolphin and abliterated Qwen have had this refusal training removed. They answer your questions directly without corporate-imposed restrictions. The model treats you like an adult.

Web search (Max only)​

When you use Max tier, the orchestrator can run a web search using the Brave Search API before sending your prompt to the worker. It fetches the top results, extracts relevant content, and includes it as context. The model then generates a response grounded in current web data and cites its sources.

This means Max tier can answer questions about recent events, look up current information, and provide sourced responses — something no browser-based tier can do.