Cheapest LLM API for coding agents.
Agents re-send big contexts. Here's what that costs.
Coding agents send large contexts (often 20,000+ input tokens per step) and produce shorter outputs. Cost per agent step (20K in, 2K out, no cache):
- GPT-5 nano: $0.0018 per step
- GPT-OSS 20B on Groq: $0.0021 per step
- Gemini 2.5 Flash-Lite: $0.0028 per step
- GPT-6 Luna: $0.0030 per step
- Claude Haiku 5.5: $0.0030 per step
- GPT-4o mini: $0.0042 per step
Among frontier models, compare Claude Opus 5.5 vs GPT-6 Astra and Claude Sonnet 5.5 vs GPT-6.1 Sol. Caching the repeated context is the biggest lever for agents.
All prices from official provider pages, checked Oct 20, 2026. See sources.
Sponsored
Your API or dev tool here. Reach developers comparing model costs right now.