Pick a winner in each dimension — change your vote anytime.
Pricing and capabilities synced from the OpenRouter catalogue. Scores are editorial (0–10).
Choose Grok 4.20 if you want agents that genuinely need 2M tokens of context. Choose Grok 4.3 if you want frontier capability on a tighter budget.
In our editorial scoring, Grok 4.3 leads in 4 of five dimensions (output quality, agentic ability, speed and value for $), while Grok 4.20 leads in 1. On price, Grok 4.20 runs about $1.63 per 1M tokens (blended) and is proprietary; Grok 4.3 is about $1.63 and proprietary.
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.