You shouldn't be running intent classification on a flagship model.
The AI Agent Action now supports Anthropic, Google and OpenAI side by side. Claude Opus 5, Sonnet 5, Haiku 4.5, Gemini 3.6 Flash, 3.1 Pro Preview and the GPT-5.6 line. Opus 5 for multi-step reasoning where
Built entirely with Claude Opus 5 and threejs
Grok 4.6 crushed Opus 5 in this BridgeBench comparison.
One benchmark is a starting signal. The useful test is how each model performs on the workflow you actually need.
I switched my main AI coding model from Opus 5 to Fable yesterday.
It's much smoother. I didn't use it much before due to cost and speed concerns. Even though the token price doubled, you can solve an issue or PR with fewer tokens. Turns out it's even?
You have reached the end of the archive
All of Claude Opus 5