Reasoning

GPT models#

Chat Completions uses reasoning_effort and the Responses API uses reasoning.effort. Typical values are low, medium and high:

from openai import OpenAI

client = OpenAI(api_key="sk-...", base_url="https://api.flyrouter.ai/v1")

resp = client.chat.completions.create(
    model="gpt-6.1-sol",
    reasoning_effort="high",
    messages=[{"role": "user", "content": "Prove that the square root of 2 is irrational"}],
)
print(resp.choices[0].message.content)

Reasoning tokens are reported in the reasoning-token breakdown of usage.

Claude models#

Newer Claude models use adaptive thinking: the model decides whether and how long to think, and you set the overall effort with output_config.effort: low, medium, high, xhigh or max.

from anthropic import Anthropic

client = Anthropic(api_key="sk-...", base_url="https://api.flyrouter.ai")

with client.messages.stream(
    model="claude-opus-5-5",
    max_tokens=32000,
    thinking={"type": "adaptive", "display": "summarized"},
    output_config={"effort": "high"},
    messages=[{"role": "user", "content": "Design a URL shortener that handles a million concurrent users. Explain the architecture and trade-offs."}],
) as stream:
    msg = stream.get_final_message()

for block in msg.content:
    if block.type == "thinking":
        print("[thinking summary]", block.thinking)
    elif block.type == "text":
        print(block.text)

• display: "summarized" returns a summary of the reasoning. Without it, thinking blocks come back with empty text (the model still thinks, and it's still billed).
• Rules of thumb: low for chat and classification; start at medium or high for everyday coding and agent work; use xhigh or max when correctness matters more than cost.

Tips#

• Thinking can take a while, so use streaming and set max_tokens high enough that the answer isn't cut off.
• In multi-turn conversations, pass the thinking blocks the model returned back unchanged in the history. Don't edit or remove them.
• Simple tasks don't need deep reasoning. Lower effort is often faster and cheaper with no loss in quality.