Reasoning Effort

reasoning_effort caps how much a model thinks before it answers. On models that reason without a bound by default it is the single largest cost lever - and what it does depends entirely on which upstream your model runs on.

Values

Send one of these on the request body. Higher effort means more reasoning tokens billed and slower replies:

minimal · low · medium · high

Example request

curl https://synthorai.io/v1/chat/completions \
  -H "Authorization: Bearer $SYNTHORAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.2",
    "messages": [{"role": "user", "content": "Refactor this function."}],
    "reasoning_effort": "low"
  }'

What happens per upstream

The gateway speaks one protocol to you and another to each provider. This table states what the parameter becomes on the way out - including the case where it is dropped.

You call	Upstream	Behaviour	說明
`/v1/chat/completions`	OpenAI-compatible (GPT, GLM, DeepSeek, Qwen, Kimi…)	透傳	Forwarded to the provider as sent, with one exception: when a model's catalog entry lists the effort levels it accepts, a value below the lowest is raised to it and one above the highest steps down (GPT-6.1 Sol cannot turn reasoning off, so none and minimal run at low). Otherwise the gateway neither validates the value nor guarantees the provider honours it. Commonly minimal, low, medium and high, but the accepted set is the provider's: OpenAI also takes none on most models, GLM takes up to max. Check the provider's own reference for its range.
`/v1/chat/completions`	Google Vertex - Gemini 4	對應	Becomes Gemini's `thinkingLevel`, kept inside the three levels Gemini 4 Argon accepts: none and minimal run at low (the model cannot turn thinking off), low/medium/high map straight across, xhigh and max run at high. An explicit `google.thinking_config` in the request wins over this mapping.
`/v1/chat/completions`	Google Vertex - Gemini 3.x	對應	Becomes Gemini's `thinkingLevel`: none and minimal both map to minimal, low/medium/high map straight across. An explicit `google.thinking_config` in the request wins over this mapping.
`/v1/chat/completions`	Google Vertex - Gemini 2.5	丟棄	Gemini 2.5 only understands `thinkingBudget` and rejects `thinkingLevel` outright, so the gateway sends neither and leaves the model at its default thinking behaviour. Use `google.thinking_config` with `thinking_budget` to control 2.5.
`/v1/chat/completions`	Anthropic (Claude)	對應	轉換為模型可接受形式的 Anthropic `thinking`。在只支援自適應思考的 Claude 型號上(Claude Opus 4.7 及以後、Sonnet 5 及以後、Haiku 5.5),low/medium/high/xhigh/max 轉為 `thinking:{"type":"adaptive"}`,並以同一檔位寫入 `output_config.effort`;minimal 依 low 處理。none 依該型號自己的寫法關閉思考:多數型號為 `{"type":"disabled"}`,Sonnet 5.5 為 `{"type":"between_tools"}`,Opus 5.5 無法關閉思考,依 low 檔位以自適應方式執行。其他取值不會轉發。在更早的 Claude 型號上則轉換為預算:none 與 minimal 關閉思考(`{"type":"disabled"}`);low/medium/high/xhigh/max 分別對應 2,048 / 8,192 / 16,384 / 32,768 token 的預算——即反方向那套分桶對應的逆對應。檔位衍生出的預算會被限制在 `max_tokens` 以內;當 `max_tokens` 連 Anthropic 的 1,024 token 下限都放不下時則完全不下發,因此設定檔位絕不會把一個原本能成功的請求變成上游 400。
`/v1/messages`	OpenAI-compatible (GPT, GLM, DeepSeek, Qwen, Kimi…)	對應	Derived from `thinking.budget_tokens` by bucket, when extended thinking is enabled and the budget is above zero: up to 2,048 tokens becomes low, up to 8,192 becomes medium, above that becomes high. A zero or absent budget sends nothing. For a model whose catalog entry lists the effort levels it accepts (GPT-6.1 Sol), the request's intent is followed instead: thinking disabled or between_tools runs at the lowest level (low), even when output_config.effort is also set; otherwise output_config.effort is sent as the effort.
`/v1/responses`	OpenAI-compatible (GPT, GLM, DeepSeek, Qwen, Kimi…)	對應	Read out of the Responses-API `reasoning.effort` object and sent as the flat `reasoning_effort` field.

Each row is transcribed from the converter that implements it and pinned by a Go test, so this table cannot drift from the running gateway without a build failing.

Cost impact

Reasoning tokens are billed as output tokens. On a model that reasons unbounded by default, setting an explicit effort is usually the difference between a few cents and a few dollars for the same answer - measure it on your own prompts before choosing a default.

Models that reason

These models document a reasoning mode. Check the model page for how its upstream handles the parameter:

ByteDance-Seed-1.8 · Dola-Seed-2.0-lite · Dola-Seed-2.0-mini · Dola-Seed-2.0-pro · claude-fable-5 · claude-fable-5-1 · claude-haiku-4-5 · claude-haiku-5-5 · claude-opus-4-5 · claude-opus-4-6 · claude-opus-4-7 · claude-opus-4-8 · claude-opus-5 · claude-opus-5-5 · claude-sonnet-4-5 · claude-sonnet-4-6 · claude-sonnet-5 · claude-sonnet-5-5 · deepseek-v4-flash · deepseek-v4-flash-0731 · deepseek-v4-pro · deepseek-v4-pro-0813 · deepseek-v4.1-flash · deepseek-v4.1-flash-1017 · gemini-2.5-flash · gemini-2.5-flash-lite · gemini-2.5-flash-native-audio · gemini-2.5-pro · gemini-3-flash-preview · gemini-3-pro-image · gemini-3-pro-image-preview · gemini-3.1-flash-image · gemini-3.1-flash-image-preview · gemini-3.1-flash-lite-image · gemini-3.1-flash-lite-preview · gemini-3.1-flash-live · gemini-3.1-pro-preview · gemini-3.5-flash · gemini-3.5-flash-lite · gemini-3.6-flash · gemini-3.7-flash · gemini-3.8-flash · gemini-4-argon · gemini-nano-banana-2.1 · glm-5 · glm-5-turbo · glm-5.1 · glm-5.2 · glm-5.3 · glm-5.3-flash · gpt-5.2 · gpt-5.2-codex · gpt-5.3-codex · gpt-5.4 · gpt-5.4-mini · gpt-5.4-nano · gpt-5.4-pro · gpt-5.5 · gpt-5.5-pro · gpt-5.6 · gpt-5.6-luna · gpt-5.6-sol · gpt-5.6-terra · gpt-6-astra · gpt-6-luna · gpt-6-sol · gpt-6.1-sol · gpt-6.1-terra · kimi-k2.5 · kimi-k2.7-code · kimi-k3 · minimax-m3 · qwen3-max · qwen3-vl-flash · qwen3-vl-plus · qwen3.5-flash · qwen3.5-plus · qwen3.6-35b-a3b · qwen3.6-flash · qwen3.7-max · qwen3.7-plus · qwen3.8-flash · qwen3.8-max

Full request body reference: Chat Completions