Tokens, never metered.Pay for parallel streams.

An OpenAI- and Anthropic-compatible inference API with no token meter. Send as much as you like — your plan sets how many generations run at once.

Get API key Read the docs

No token meter
OpenAI & Anthropic SDKs
Built for agents

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.nevermetered.com/v1",
    api_key=os.environ["NEVERMETERED_API_KEY"],
)
stream = client.chat.completions.create(
    model="MODEL_ID",
    messages=[{"role": "user",
               "content": "Why streams?"}],
    stream=True,
)
for chunk in stream:
    delta = chunk.choices[0].delta
    print(delta.content or "", end="")

streaming · stream 3 of 8 · 1,842 cached input tokens

Parallel streams let your agents run side by side instead of in single file. Each generation uses one of your plan's slots, and repeated prompt prefixes can be served from cache.

Works with your tools

Why teams switch

Built for agents, pipelines and products that stream all day

Per-token pricing punishes the workloads that matter most. We sell capacity instead — predictable, dedicated, and fast.

No token meter

No per-token pricing and no monthly token quota. Send long contexts and long answers — your plan price never moves.

Dedicated parallel streams

You pick how many generations run at once, and give each key its own cap — a runaway job can't use more than its share of your streams.

Fair scheduling

When the pool is busy, requests queue fairly: users with lighter recent usage go first, and waiting requests gain priority over time.

OpenAI + Anthropic compatible

Chat Completions, Responses and Messages on one base URL. Point the official SDKs at it and keep your code.

Prefix caching, sticky sessions

Conversations stick to the same engine for a while, so repeated prompt prefixes can be served from cache instead of recomputed.

Full visibility

Every request logged with input, cached input and output tokens, latency and GPU time. Export anything to CSV.

How it works

Live in three steps

If your code already talks to OpenAI or Anthropic, switching is a base URL and a key.

1
Create a key
Sign up and generate an API key in seconds. Optionally cap how many streams each key may use.
2
Point your SDK
Swap the base URL in your OpenAI or Anthropic client. Your prompts, tools and streaming code stay the same.
3
Scale with streams
Run as many generations as your plan allows, in parallel. Need more? Upgrade — tokens stay unlimited.

Pricing

Pay for parallelism, not tokens

Every plan includes unlimited input and output tokens. Choose how many streams you need at once.

Plans are being finalized

Create a free account now — every account starts with a free daily allowance and unlimited tokens per request.

FAQ

Questions, answered

What does “never metered” actually mean?

There is no per-token charge and no monthly token quota. Your plan sets how many requests can generate at the same time (parallel streams) and a per-request output cap. Everything you send and receive within that is included.

Is there a free tier?

Yes. Every account can send a small number of requests a day for free once its email address is verified, and the allowance resets at 00:00 UTC. Free requests are shared per device and network, so one person gets one allowance. Paid plans have no daily request limit.

What happens when I use all my parallel streams?

Extra requests wait in a queue, up to your plan’s queue allowance, and start when one of your streams frees up. Beyond that the API returns 429 concurrency_limit_exceeded right away, so your client can back off.

Which APIs are supported?

OpenAI Chat Completions and Responses, and Anthropic Messages (including token counting), all on the same base URL. Use the official SDKs by changing the base URL and API key.

How do cached input tokens work?

Requests in the same conversation are routed to the same engine for a while, so a prompt prefix it has already processed can be served from its cache. Cache hits show up as cached input tokens in your usage; how much they help depends on how stable your prompts are.

Can I see exactly what I used?

Yes. The dashboard shows every request with input, cached input and output tokens, latency and status, plus daily rollups by key and model — all exportable as CSV.

Can I limit what each API key can do?

Each key can have its own maximum number of parallel streams, so a background job can never crowd out your production traffic.

Do you store or train on my prompts?

No. Prompts and completions are processed in memory on our inference servers and are never written to our database or logs, and we never use your content to train models. We keep request metadata such as token counts, timings and status, which is what powers your usage dashboard. Details are in the Privacy Policy.

Stop counting tokens. Start shipping.