AI model routing
Kimi K3, DeepSeek v4.1 Flash 1017, and Meta Muse Spark Open. Use one OpenAI-compatible endpoint with clear pricing and usage receipts.
For specialized workloads, we run custom evals, train models, and manage serving. See custom models →
Change the base URL. Keep your SDK.
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["LOGITS_API_KEY"],
base_url="https://api.thinwire.ai/v1",
)
response = client.chat.completions.create(
model="deepseek-v4.1-flash",
reasoning_effort="low",
max_tokens=2048,
messages=[{"role": "user", "content": "Hello"}],
)Live request path
Example requestmodel routing
Pay for the models you use. DeepSeek and Muse use paid credits at the rates shown below.
models
Explore Kimi K3, DeepSeek, and Muse with transparent token pricing and rate limits.
kimi-k3
1,048,576-token model context with native token counting; text prompts are admitted up to 4 MiB of request bytes. Reasoning, streaming, and tool calls.
Also available as kimi-k3-base. Uses paid or available promotional credits.
deepseek-v4.1-flash
Serving the DeepSeek-V4.1-Flash-1017 snapshot (released 2026-10-17). 1,048,576-token model context with native token counting; text prompts are admitted up to 4 MiB of request bytes. Reasoning, streaming, tool calls, and logprobs.
Also answers to deepseek-v4.1-flash-1017 and deepseek-v4.1-1017; prices and receipts stay on deepseek-v4.1-flash. Uses paid or available promotional credits.
muse-spark-open
50% offOpen-weight model optimized for agentic coding and tool calling.
Text API rate limits apply per user, per model, over a rolling 60-second window. If you receive HTTP 429, wait for the Retry-After interval before retrying.
next
One API for model routing, plus custom evals and model serving.
connect