MyAi vs OpenAI
Drop-in OpenAI-compatible API powered by community GPU operators. Migration is one env var:
OPENAI_BASE_URL="https://api.myaitoken.io/v1" OPENAI_API_KEY="myai-sk-..." # free key auto-issued on first call via the MCP
Side-by-side
| Capability | MyAi | OpenAI |
|---|---|---|
| API compatibility | OpenAI-compatible /v1/chat/completions, /v1/models | Native |
| Models | 22 open-weight (qwen3.6:27b, qwen2.5-coder:32b, deepseek-r1, llama3.2-vision, more) | GPT-4o, GPT-4.1, o1, o3 (closed) |
| Pricing | Free tier (100 free completions/machine, no card) + per-token paid in $MYAI or USD | $2.50/$10 per M tokens (GPT-4o), no free tier on production |
| Where inference runs | Community GPU operators worldwide — operators earn $MYAI per completion | Microsoft Azure datacenters |
| Data policy | Operators run locally; never leaves their box; no central log | Retained 30 days by default; opt-out via API account settings |
| Censorship resistance | Decentralized supply; no single party can revoke access | Single vendor; rate limits + content policy enforced centrally |
| Streaming | SSE, server-side | SSE, server-side |
| Function calling / tools | Yes (OpenAI-format) | Yes (native) |
| Vision | Yes via llama3.2-vision, minicpm-v | Yes via GPT-4o |
| Embeddings | Yes (nomic-embed-text, mxbai-embed-large) | Yes (text-embedding-3-*) |
| Latency (p50, cache hit) | <200ms | <200ms |
| Latency (p99, cold) | ~3-5s (network-dependent) | <1s |
| Top-up | Buy $MYAI on Aerodrome (Base) or USD via Stripe | Credit card via dashboard |
| Best for | Teams that want open-weight models, decentralization, and lowest-cost inference | Teams that need frontier closed models (GPT-5, o3) and don't care about open weights |
When OpenAI is the right call
- You need a specific closed model (GPT-5, o3, the new Vision SOTA).
- You're already invested in the OpenAI Realtime / Assistants APIs.
- You require centralized SLAs that a decentralized supply can't yet match.
When MyAi is the right call
- You want the best open-weight models without spinning up your own GPUs.
- You're building an agent that needs censorship-resistant inference.
- You want pay-per-completion in crypto-native rails (no Stripe required).
- You're cost-sensitive — community GPUs underprice cloud inference 5–10×.
Try it in 30 seconds
No signup. Free key auto-issued on first call via the MCP:
uvx myai-mcp
Or use the OpenAI SDK directly with our base URL — same calls, different network.