OpenAI o-series

Reasoning models, represented by OpenAI o1, DeepSeek-R1, and QwQ, have chain-of-thought expansion as their core mechanism, generating a large number of intermediate reasoning steps before producing a final answer.

推理能力基于思维链（Chain-of-Thought, CoT）技术，模型在给出最终答案前生成内部逐步推导过程

Google's Gemini, Anthropic's Claude, and OpenAI's o-series models have all implemented different forms of extended thinking.

OpenAI's o-series reasoning models introduce a slow-thinking mechanism that performs internal chain-of-thought reasoning before outputting the final answer.

Reasoning Effort is essentially a parameter that controls the depth of the model's chain of thought; the higher the level, the more internal reasoning steps the model is allowed, and latency and cost rise proportionally.

Reasoning models, represented by OpenAI's o series, DeepSeek-R1, and the Kimi K series, perform a chain-of-thought process before outputting, decomposing problems into multiple steps, performing self-verification, and deriving answers incrementally.

The overthinking problem is not unique to Qwen3; reasoning models such as DeepSeek-R1 and the OpenAI o1 series all exhibit similar phenomena to varying degrees.

Reasoning tokens are the internal thinking content the model generates before giving its final answer, usually billed at the output rate, and can account for an extremely high share of total Agent output (77% of the bill in the example).

In extended thinking mode, Claude counts thinking tokens and final output tokens against the same max_tokens budget, whereas OpenAI's o-series models have a separate budget cap for reasoning tokens.