Pick a winner in each dimension — change your vote anytime.
Pricing and capabilities synced from the OpenRouter catalogue. Scores are editorial (0–10).
Choose Claude Haiku 4.5 if you want cheap, fast subtasks and subagents inside a larger workflow. Choose Qwen3 235B A22B if you want self-hosted or ultra-cheap open-weight agents.
In our editorial scoring, Claude Haiku 4.5 leads in 4 of five dimensions (output quality, agentic ability, speed and reliability), while Qwen3 235B A22B leads in 1. On price, Claude Haiku 4.5 runs about $2.20 per 1M tokens (blended) and is proprietary; Qwen3 235B A22B is about $0.09 and open-weight.
The model picks the moves; the agent runs the loop, the tools, and the guardrails. Once you've chosen a model, see which agent gets the most out of it.