Flynn preview: your AI tools on open-weight models, without the rewrite

Open-weight models are good enough for a lot of real work now. Moving to them is still a chore: every tool expects a particular provider, and the models themselves trip up agents in small ways.

Flynn is our answer, and it is now in preview. You keep the tools and code you already have, and they run on open-weight models. We're opening it up gradually: request early access and we'll get you set up.

Keep the tools you use

Claude Code, the OpenAI SDK and the Anthropic SDK work as they are. You change the address and the key, nothing else. For Claude Code that's two settings:

export ANTHROPIC_BASE_URL=https://api.flynn.digitalfrontier.so
export ANTHROPIC_AUTH_TOKEN="$FLYNN_API_KEY"
claude

Big jobs in one go

Evaluations, labelling, bulk generation: send up to 10,000 requests as one batch and collect the results within a 24-hour window. Use OpenAI's batch format or Anthropic's, whichever your code already speaks, and read the results in either. Anything that can't run is listed with the reason, never lost.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.flynn.digitalfrontier.so/v1",
    api_key=os.environ["FLYNN_API_KEY"],
)

upload = client.files.create(file=open("requests.jsonl", "rb"), purpose="batch")
batch = client.batches.create(
    input_file_id=upload.id,
    endpoint="/v1/chat/completions",
    completion_window="24h",
)

The right model for each task

Ask for the model flynn and let Flynn choose. It learns which model handles each kind of task best for what it costs, and keeps learning as it goes. If you would rather choose yourself, name any model Flynn serves.

Fewer broken agent runs

Open models often get tool calls slightly wrong: a stray character, a value in the wrong shape. Agents that check tool calls, as Claude Code does, turn many of those into errors and lose a turn. Flynn fixes the slips it can prove are safe, and leaves everything else exactly as the model wrote it. It never makes up a value.

Try it

1. Request early access.
2. Point Claude Code, your OpenAI or Anthropic code, or your batch job at https://api.flynn.digitalfrontier.so.

You pay per token, and failed requests are not billed. Prices are on the Flynn pricing page.

Read more

• Flynn product page
• Introduction and API reference
• Flynn Service Terms

Questions? Write to support@digitalfrontier.so.