Reasoning Model Integration

Step Plan supports accessing StepFun reasoning models via a dedicated path. All requests uniformly use the /step_plan/v1/... path prefix, and the domain name is fixed as https://api.stepfun.ai.

Prerequisites

1. Subscribed to a Step Plan.
2. Obtained an API Key.

---

Supported Models

Model	Description
step-3.7-flash	Flagship multimodal reasoning model with native image and video understanding. Supports three reasoning effort levels (low/medium/high) — well-suited to agent, coding, and multimodal workloads.
step-3.5-flash-2603	Optimized from Step 3.5 Flash for high-frequency Agent scenarios. Improved token efficiency and faster inference; can switch to low-inference mode to significantly reduce token consumption.
step-3.5-flash	Sparse MoE architecture with 196B total / 11B activated parameters. High-speed inference, optimized for agent and coding tasks.

Endpoint Paths

Capability	Request Method	Step Plan Path
Chat Completion (OpenAI protocol)	POST	https://api.stepfun.ai/step_plan/v1/chat/completions
Messages (Anthropic protocol)	POST	https://api.stepfun.ai/step_plan/v1/messages

The endpoint parameters are exactly the same as the open platform. For details, see the Chat Completion API docs and the Messages API docs.

The Anthropic SDK automatically appends /v1/messages to the base URL, so when using the Anthropic SDK, set the base URL to https://api.stepfun.ai/step_plan (without /v1). The OpenAI SDK uses https://api.stepfun.ai/step_plan/v1.

Reasoning Effort

Models that support three reasoning effort levels accept low, medium, or high in the request. Chat Completion (OpenAI protocol) uses reasoning_effort; Messages (Anthropic protocol) uses output_config.effort.

Effort	Recommended Use
low	Simple Q&A, summarization, rewriting, information extraction
medium	Default recommendation; suits general reasoning and multi-step tasks
high	Complex reasoning, math, planning, code analysis

Billing

The billing logic is consistent with the open platform. The actual amount billed on the open platform is converted into Step Plan total quota consumption. For details on plan entitlements, see the Step Plan overview.

Integration Methods

Direct API Calls

curl

curl -X POST 'https://api.stepfun.ai/step_plan/v1/chat/completions' \
-H 'Content-Type: application/json' \
-H "Authorization: Bearer $STEP_API_KEY" \
-d '{
    "model": "step-3.7-flash",
    "messages": [
        {"role": "user", "content": "Hello, please introduce yourself."}
    ]
}'

Python (OpenAI SDK)

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_STEP_API_KEY",
    base_url="https://api.stepfun.ai/step_plan/v1",
)

response = client.chat.completions.create(
    model="step-3.7-flash",
    messages=[
        {"role": "user", "content": "Hello, please introduce yourself."}
    ],
)

print(response.choices[0].message.content)

Python (Anthropic SDK)

from anthropic import Anthropic

# Note: The Anthropic SDK automatically appends /v1/messages; base_url should not include /v1
client = Anthropic(
    api_key="YOUR_STEP_API_KEY",
    base_url="https://api.stepfun.ai/step_plan",
)

message = client.messages.create(
    model="step-3.7-flash",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Hello, please introduce yourself."}
    ],
)

print(message.content[0].text)

Via Tool Integrations

Reasoning models can be integrated through a variety of Agent tools and coding assistants. Just set the Base URL to https://api.stepfun.ai/step_plan/v1 and select step-3.7-flash, step-3.5-flash-2603, or step-3.5-flash as the model.

See the Quick Start and the individual tool integration guides:

OpenClaw

Command-driven Agents and initialization-based workflows.

Claude Code

Coding, debugging, and engineering collaboration in the terminal.

Hermes-Agent

Open-source AI Agent framework for terminals or messaging platforms.

Open Code

Drive development tasks in the terminal with natural language.