// everyday driver
Sonnet 4.6 vs Gemini 3 Pro
Two models built for the boring-but-critical middle: RAG, chat, batch inference, structured output. Different DNA, different tradeoffs.
Sonnet 4.6
Gemini 3 Pro
Context
200K
1M
Input / M (llmrelay)
$1.50
$0.625
Output / M (llmrelay)
$7.50
$5.00
Multimodal
Text + image
Text + image + video
Tool use
Excellent
Good
Latency
~2× faster
Baseline
Pick Sonnet 4.6 when
You care about tool use, structured output, and API polish. Sonnet is 4× more expensive per input token but the delta closes fast on tasks under 100K context — Anthropic's tool-call reliability is worth it.
Pick Gemini 3 Pro when
You need 1M context (huge codebase understanding, video summarization, multi-document RAG) or you need multimodal video. Also just when the bill would be too big on Sonnet at scale.