OpenCode
Terminal-first coding agent with 75+ providers
OpenCode is an MIT-licensed terminal AI coding agent with a desktop app and IDE extension. It supports 75+ LLM providers through Models.dev, local models, and hot-swapping between providers without restarting your session.
Command Code
Agent that learns your coding taste, with model switching built in
Command Code is a terminal coding agent focused on open models and per-project 'taste' learning. It exposes every supported model through one CLI, with per-model credit allowances and zero-markup pay-as-you-go top-ups.
Pi
Minimal, hackable agent harness from Earendil
Pi is an MIT-licensed minimal coding-agent harness. It ships four modes (interactive, print/JSON, RPC, SDK), supports 15+ providers, and is designed to be extended with TypeScript packages rather than configured through a large feature matrix.
OpenRouter
One API key, 400+ models, automatic provider routing
OpenRouter exposes a single OpenAI-compatible endpoint with per-model pricing, provider routing, fallbacks and usage accounting. It is the simplest way to compare models without opening accounts with every lab.
Why Switch Models at Peak?
DeepSeek charges peak rates on weekdays from 01:00–04:00 and 06:00–10:00 UTC, and 50% less during every other hour, including the full weekend. Off-peak DeepSeek V4.1 Flash costs $0.15/M input, $0.003/M cache hits and $0.60/M output; peak doubles those rates. Route long refactors, batch test loops and overnight agents into off-peak windows, or fall back to open-weight models with flat pricing such as GLM-5.3 Flash or Kimi K3 when you need premium throughput.