API

API code generator

Pick a language and a task to get code that follows Reflection's documentation, including the details that differ from OpenAI.

Beam-501B-A23B

Language

Task

reasoning_effort

max_completion_tokensEmpty means the API default. Reasoning counts toward this limit.

System prompt

User message

Print the reasoning

Install and key

bash

pip install openai
export REFLECTION_API_KEY="<your API key>"

Code

main.py

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.reflection.ai/openai/v1",
    api_key=os.environ["REFLECTION_API_KEY"],
)

messages = [
    {"role": "user", "content": "Write a short poem about sparse experts."},
]

stream = client.chat.completions.create(
    model="Beam-501B-A23B",
    messages=messages,
    reasoning_effort="medium",
    stream=True,
    stream_options={"include_usage": True},
)

for chunk in stream:
    if not chunk.choices:  # the last chunk carries only usage
        print("\n", chunk.usage)
        continue
    delta = chunk.choices[0].delta
    reasoning = getattr(delta, "reasoning_content", None)
    if reasoning:
        print(reasoning, end="", flush=True)
    if delta.content:
        print(delta.content, end="", flush=True)

Reasoning streams first in delta.reasoning_content, then the answer in delta.content. include_usage adds a final chunk with token counts.

Parameter support

Supported

model, messages with system, developer, user, assistant and tool roles
reasoning_effort: low, medium, high, xhigh, max
stream and stream_options.include_usage
tools, tool_choice, parallel_tool_calls
response_format: json_object and json_schema, with strict
temperature, top_p, frequency_penalty, presence_penalty, seed, max_completion_tokens

Not supported

Image, audio and file inputs
n other than 1, logprobs, logit_bias
user, metadata, audio and prediction return an error
stop is accepted but has no effect
Responses, Embeddings, Images, Audio, Files, Batch and Assistants APIs

Errors worth handling

Status and code	What it means
400 unsupported_value	A reasoning_effort value the model doesn't accept, such as none or minimal.
400 missing_required_parameter	An earlier assistant message was sent without its reasoning_content, or json_schema is missing.
401	The API key is missing or invalid.
429 rate_limit_exceeded	A per-minute or daily limit was hit. Wait for Retry-After.
503 inference_capacity_unavailable	Capacity is temporarily exhausted. Retry after Retry-After.

Sources: openai-compatibility · reasoning · tool-calling · structured-outputs · errors