#Halp
1 messages · Page 1 of 1 (latest)
Short version:
- Haiku 4.5 = better raw cost efficiency for simple stuff, fast replies, classification, light extraction, cheap background tasks.
- Sonnet 4.6 = better ability per request, especially for coding, tool use, multi-step reasoning, and “get it right the first time” work.
So the practical answer is:
- If you want the cheapest decent model, pick Haiku
- If you want the best balance for a main agent, pick Sonnet
- If the work is agentic and mistakes/retries are expensive, Sonnet is often more cost-effective in practice even if the per-token price is higher
My rule of thumb for OpenClaw:
- Main assistant / coding / tools → Sonnet
- Heartbeats / triage / lightweight chores → Haiku
For current community recommendations, check:
- Models docs: https://docs.openclaw.ai/concepts/models