What is a Reasoning Model?
A Reasoning Model is an LLM that “thinks before it answers.” Traditional LLMs stream the answer token by token; reasoning models first generate a long internal chain-of-thought — decomposing the problem, trying approaches, backtracking, verifying — and only then return the final answer to the user. OpenAI o1/o3, Claude Opus Extended Thinking, Gemini Thinking, and DeepSeek R1 all belong to this class.
Hands-on lesson: complex trading strategy derivation, multi-step debugging, math proofs, and any agent task that needs planning go much better on reasoning models than on vanilla LLMs. The cost is slower responses, much higher token usage, and pricier API calls. We reserve Opus + Extended Thinking for COO-level work and trading research, and use Sonnet or Haiku for daily chat and summaries.