---
title: "Flat-rate vs pay-per-token LLM API - break-even math"
description: "When does a flat-rate LLM plan beat paying per token? Four worked monthly bills for DeepSeek V4.1 Flash, the break-even volume, and how to measure your own."
url: https://broadrouter.com/blog/flat-rate-vs-pay-per-token-llm-api/
lang: en
updated: 2026-11-14
publisher: BroadRouter (a BroadNet site); API by BroadNet Inference, https://inference.broadnet.ai
---

# Flat-rate vs pay-per-token LLM API: the break-even math

By the BroadNet Inference team · 8 min read · Published 2026-11-14 · Updated 2026-11-15

For DeepSeek V4.1 Flash on BroadNet Inference, a flat Full Moon plan ($60 a month) beats pay-per-token at roughly 917 million input and 46 million output tokens a month, assuming 80% cache hits. A full-time coding agent usually crosses that line. A side project or chat assistant usually does not.

## Key takeaways

-   Pay per token costs $0.175 per 1M input, $0.0055 per 1M cached input and $0.52 per 1M output for DeepSeek V4.1 Flash.
-   Full Moon is $60 per session per month with no token meter; one session runs one request at a time.
-   Break-even at 80% cache hits and 20 input tokens per output token: about 917 million input plus 46 million output tokens a month.
-   A full-time coding agent in our example costs $65.40 a month per token, so the plan wins; a daily assistant costs $42.67, so per-token wins.
-   Measure a week of real usage, including `cached_tokens`, before you choose.

**On this page**

1.  What are the two ways to pay?
2.  How do you calculate the break-even point?
3.  Which workloads are cheaper on a flat plan?
4.  Where does the break-even line fall?
5.  Flat rate or per token: side by side
6.  What moves the break-even point?
7.  Who else sells flat-rate LLM API access?
8.  What mistakes skew the comparison?
9.  How do you measure your own usage before choosing?
10.  Where BroadNet Inference fits

A flat-rate LLM plan is cheaper than paying per token only above a break-even volume. For DeepSeek V4.1 Flash on BroadNet Inference, the $60 Full Moon plan breaks even at about **917 million input and 46 million output tokens a month** with 80% cache hits. A full-time coding agent usually crosses that line. A side project or chat assistant usually does not.

Below are the formula, four worked monthly bills and the rival flat-rate plans, all from published prices. You can redo the math with your own numbers in the [pricing calculator](https://broadrouter.com/pricing/).

## What are the two ways to pay?

[Pay-as-you-go](https://broadrouter.com/glossary/pay-as-you-go/) prices for [DeepSeek V4.1 Flash](https://broadrouter.com/models/deepseek-v4.1-flash/), per 1 million tokens:

| Token type | Price per 1M |
| --- | --- |
| Input | $0.175 |
| [Cached input](https://broadrouter.com/glossary/cached-input-tokens/) | $0.0055 |
| Output (includes reasoning) | $0.52 |

The [flat-rate plans](https://broadrouter.com/glossary/flat-rate-llm-plan/) cover the same model with no token meter:

| Plan | Price per session | Period |
| --- | --- | --- |
| Full Moon | $60 | Month, renews |
| Orbit | $15 | Week |
| Meteor | $1 | Hour |

A session runs **one request at a time**. That is the only limit: no monthly token cap, no rolling window and no weekly quota. [How sessions work](https://broadrouter.com/guides/unlimited-llm-api-sessions/).

## How do you calculate the break-even point?

The math has three steps. You need your own token mix, not a list price.

1.  **Find your blended price per 1M tokens.** Add up a month of tokens by type and multiply each by its price: `cost = new input × $0.175 + cached input × $0.0055 + output × $0.52` (all in millions of tokens). Divide by the total tokens.
2.  **Divide the plan price by that blend.** The result is how many million tokens a month the plan replaces.
3.  **Compare with your real volume.** Above the line, the plan is cheaper. Below it, pay per token.

Here is the same math on one agent request, so you can see where the money goes. Say a coding agent sends a 50,000-token prompt, 40,000 of it read from cache, and gets 1,500 tokens back:

| Part | Tokens | Billed at |
| --- | --- | --- |
| New input | 10,000 | $0.175 per 1M |
| Cached input | 40,000 | $0.0055 per 1M |
| Output and reasoning | 1,500 | $0.52 per 1M |

Multiply each row by its price and divide by 1,000,000. Then multiply by the number of requests a day. One request costs a fraction of a cent, but an agent that makes a few hundred calls a day turns that into a real monthly bill. The [flat-rate vs per-token calculator](https://broadrouter.com/tools/flat-rate-vs-per-token-calculator/) does this for you.

## Which workloads are cheaper on a flat plan?

These examples use the published prices above. The cache share is the fraction of input tokens read from cache, which is high for agents that resend the same files and system prompt on every turn.

| Workload | Input / month | Cache hits | Output / month | Pay per token | Cheaper option |
| --- | --- | --- | --- | --- | --- |
| Side project | 100M | 50% | 10M | $14.23 | pay as you go |
| Daily assistant | 300M | 50% | 30M | $42.67 | pay as you go |
| Full-time coding agent | 1,000M | 80% | 50M | $65.40 | Full Moon |
| Always-on agent | 3,000M | 80% | 150M | $196.20 | Full Moon |

The difference against one Full Moon session ($60) is $45.77, $17.33, $5.40 and $136.20 a month respectively. The always-on agent is the clear case: the plan bill stays the same however much the agent reads.

## Where does the break-even line fall?

Coding agents read far more than they write. A typical shape is about 20 input tokens for every output token, with most of the input cached. At that mix, one million tokens cost a blended $0.06. The plan pays for itself at about **917 million input and 46 million output tokens a month**.

One plan session also has a ceiling. At a typical speed of about 100 tokens per second, a session that generated non-stop for 30 days would produce about 259 million output tokens. Real agents spend much of their time reading, running tools and waiting, so the practical number is lower. If one agent is not enough, add sessions; see [running parallel agents on a flat rate](https://broadrouter.com/blog/run-parallel-coding-agents-flat-rate/).

## Flat rate or per token: side by side

|  | Pay per token | Flat-rate plan session |
| --- | --- | --- |
| Bill grows with | Every token, retry and long prompt | Number of sessions only |
| Best for | Light, spiky or experimental use | Steady agents that run for hours |
| Parallel requests | 4 in flight, 30 a minute per key | One per session; add sessions to add parallelism |
| Models | Every chat and speech model | DeepSeek V4.1 Flash only |
| Cache rate matters | Yes, a lot | No, for the bill |
| Smallest commitment | Nothing beyond your balance | One $1 Meteor hour |

There is also a third option. An hourly key for [BroadNet Gemma 4 VL 26B](https://broadrouter.com/models/gemma-4-vl-26b/) costs $0.50 an hour, with one stream and 120 requests a minute. It suits short, heavy jobs on a small vision model rather than all-day agents; running one around the clock for 30 days would cost $360.00.

## What moves the break-even point?

-   **Cache rate.** Cached input costs $0.0055 per 1M instead of $0.175. A higher cache rate makes pay-per-token cheaper and moves the break-even point up. [How to raise your cache rate](https://broadrouter.com/blog/cached-input-tokens-explained/).
-   **Reasoning.** DeepSeek V4.1 Flash thinks before it answers, and those [reasoning tokens](https://broadrouter.com/glossary/reasoning-tokens/) are billed as output. Set `reasoning_effort` to `low` for simple steps. [Why reasoning can empty a reply](https://broadrouter.com/guides/reasoning-tokens-empty-response/).
-   **Concurrency.** A pay-as-you-go key allows 4 requests in flight and 30 a minute. A plan allows one request per session, but you can buy as many sessions as you run agents.
-   **Where you buy per-token access.** DeepSeek’s own API also sells V4.1 Flash per token, with peak and off-peak rates. On 15 November 2026 its off-peak input prices (USD 0.15 per 1M, or USD 0.003 cached) were lower than BroadNet’s, and its output price (USD 0.60 off-peak, USD 1.20 peak) was higher at all hours, so the cheaper per-token option depends on your mix and time of day. Several other providers list the model below BroadNet’s per-token price too: see the [DeepSeek V4.1 Flash price comparison across providers](https://broadrouter.com/llm-api-pricing/deepseek-v4.1-flash/). If you stay on per-token billing, shop on price; the plan is what those providers don’t sell.
-   **Predictability.** A per-token bill grows with every retry loop and every extra file. A flat bill does not, which is often worth something by itself for teams with a fixed budget.

## Who else sells flat-rate LLM API access?

BroadNet is not the only flat-price option. These are the rival plans we could verify on each provider’s own pricing page on 15 November 2026. Prices are per month. Check the allowed-use terms as well as the price: one of them rules out agents.

| Provider and plan | Price | Concurrency | Models and context | What “flat” means |
| --- | --- | --- | --- | --- |
| Featherless Chat | USD 25 | 4 concurrent units (“one 70B conversation, or four small ones at once”) | Any model in its catalogue, 32K context | “Unlimited tokens”, but only for “human-driven use”: the page rules out “app/API traffic”, “background automation” and coding agents |
| Awan LLM Core / Plus / Pro / Max | USD 5 / 10 / 20 / 80 | Not stated; Pro and Max list “parallel requests” | Small, medium and large model tiers; context not stated | “Unlimited tokens”, but capped at 20 to 200 requests a minute and, below Max, a daily request quota per model size |
| Standard Compute Starter / Economy / Standard / Pro | USD 19 / 39 / 89 / 249 | Starter: one agent at a time; Economy and up: parallel agents | Its full catalogue, routed per request | A monthly compute budget equal to the plan price; requests stop when it runs out |
| BroadNet Full Moon | $60 per session | One request per session; add sessions for more | DeepSeek V4.1 Flash only, 1,048,576-token context | No token meter, no daily or weekly cap |

**Where they beat BroadNet.** Awan LLM is far cheaper to start with, offers more models and lists parallel requests on its Pro and Max plans. Standard Compute routes to frontier models that BroadNet does not serve. Featherless Chat is cheaper and gives four concurrent units across a large catalogue, but only for chat you type yourself, so it does not fit an agent workload. If you need model choice or several parallel streams on one subscription, test Awan or Standard Compute first.

**Where BroadNet differs.** Its plan has no request-per-day quota (unlike Awan below Max) and no spend budget that runs out (unlike Standard Compute). It allows agents and API traffic, which Featherless Chat does not, and a far longer context than Featherless Chat’s 32K, which matters for coding agents that resend whole files. Whether a plan pays off still comes down to the break-even math above, so run your own numbers in the [flat-rate vs per-token calculator](https://broadrouter.com/tools/flat-rate-vs-per-token-calculator/).

## What mistakes skew the comparison?

-   **Comparing against a chat subscription.** A chat app plan is not an API key you can point your own tools at. Compare API prices with API plans. [What “unlimited” really covers](https://broadrouter.com/blog/what-unlimited-llm-api-really-means/).
-   **Ignoring cached tokens.** Pricing every input token at the full rate makes pay-per-token look far more expensive than it is.
-   **Measuring a quiet week.** A week of holidays or a week of setup is not a normal month. Measure a week of real work.
-   **Forgetting parallelism.** Three agents running at once need three sessions. Compare three plans with the per-token bill for all three.
-   **Forgetting the founders offer.** With a founders code, Full Moon is $45 a month for the first three months, which lowers the break-even point for that period.

## How do you measure your own usage before choosing?

1.  Run your real agent for a week on a pay-as-you-go key. New accounts start with $5 of credit.
2.  Turn on usage in streams with `stream_options: {"include_usage": true}` and log `prompt_tokens`, `prompt_tokens_details.cached_tokens` and `completion_tokens`. The [streaming guide](https://broadrouter.com/guides/streaming-chat-completions/) shows how.
3.  Multiply the week by about 4.3 and put the totals into the [calculator](https://broadrouter.com/pricing/).
4.  If the plan wins, buy a $1 Meteor hour and run the same loop on it before committing to a month.

## Where BroadNet Inference fits

BroadNet sells both models of billing on one account, so you do not have to choose once and for all. Steady agents on DeepSeek V4.1 Flash fit a plan session. Other models, speech to text and occasional bursts stay on pay-as-you-go. If your volume sits well below the break-even line, the plan is the wrong choice, and the numbers above will show it. The calculator and this post use the same price file, [pricing.json](https://inference.broadnet.ai/pricing.json?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-flat-rate-vs-pay-per-token-llm-api&utm_content=body), last synced 2026-11-14.

Free tools for this

-   [LLM API cost calculator](https://broadrouter.com/tools/llm-api-cost-calculator/) Price a workload on every model at once, with cache hits, reasoning and a flat-plan check.
-   [Flat-rate vs per-token LLM cost calculator](https://broadrouter.com/tools/flat-rate-vs-per-token-calculator/) Break-even between an unlimited plan and paying per token, per agent.
-   [Prompt caching savings calculator](https://broadrouter.com/tools/prompt-caching-calculator/) Bill with and without caching, by prefix share and hit rate.
-   [Prompt token diff](https://broadrouter.com/tools/prompt-token-diff/) Two prompt versions side by side: changed lines, exact token delta and the monthly cost difference.

[All free tools](https://broadrouter.com/tools/)

## Terms in this article

[Token](https://broadrouter.com/glossary/token/)

A token is the unit of text a language model reads and writes: a word, part of a word, a punctuation mark or a byte sequence produced by the model's tokenizer. LLM APIs count, limit and usually bill usage in tokens.

[Pay-as-you-go](https://broadrouter.com/glossary/pay-as-you-go/)

Pay-as-you-go is a pricing model in which you pay only for what you use, metered per unit such as tokens, requests or audio minutes, with no subscription or minimum commitment. Most LLM APIs bill this way, from a prepaid balance or a periodic invoice.

[Concurrency](https://broadrouter.com/glossary/concurrency/)

Concurrency is the number of requests a client has in flight at the same time. LLM APIs often cap it per key or plan, separately from requests per minute, because each open request holds GPU capacity until its response has finished.

[AI coding agent](https://broadrouter.com/glossary/coding-agent/)

An AI coding agent is a software tool in which a language model works on a codebase in a loop: it reads files, edits them, runs commands such as tests, reads the output and decides the next step until the task is done or a limit is reached.

[Input and output tokens](https://broadrouter.com/glossary/input-and-output-tokens/)

Input tokens are the tokens a language model reads in a request, including the system prompt, conversation history, tool definitions and images. Output tokens are the tokens it generates in reply, including any reasoning. LLM APIs usually price them separately, and output usually costs more.

[Tool calling](https://broadrouter.com/glossary/tool-calling/)

Tool calling, also called function calling, is an LLM API feature in which the model returns a structured request to run a function the developer has defined, with JSON arguments. The application runs the function and sends the result back; the model never runs code itself.

[Cached input tokens](https://broadrouter.com/glossary/cached-input-tokens/)

Cached input tokens are prompt tokens that an LLM provider has recently processed and can reuse from a cache instead of computing again, usually a repeated prefix such as a system prompt or document. Providers bill them at a lower rate than new input tokens.

[Rate limit](https://broadrouter.com/glossary/rate-limit/)

A rate limit is a cap an API places on how much a client can use it in a time window, such as requests per minute or tokens per minute. Requests over the limit are refused, usually with HTTP 429 Too Many Requests, until usage falls back under it.

## Questions

### Is a flat-rate LLM plan always cheaper?

No. Below the break-even volume, paying per token is cheaper. In our examples a side project costs $14.23 a month per token, well under the $60 plan.

### How do I calculate the break-even point for an LLM subscription?

Work out your blended cost per 1M tokens from your real mix of new input, cached input and output. Divide the plan price by that number. For DeepSeek V4.1 Flash at 80% cache hits and 20 input tokens per output token, the blend is $0.06 per 1M, so Full Moon breaks even at about 917 million input plus 46 million output tokens a month.

### Do cached tokens change the answer?

A lot. DeepSeek V4.1 Flash bills cached input at $0.0055 per 1M tokens instead of $0.175. A high cache rate pushes the break-even point higher, so measure it first.

### What if I run several agents at once?

Each plan session runs one request at a time, so you need one session per agent working in parallel. Compare the per-token bill for all agents with one plan per agent.

### Are reasoning tokens billed on a flat-rate plan?

Not separately. On pay-as-you-go, reasoning tokens are billed as output. On a plan session there is no token meter, so reasoning, retries and long prompts do not change the bill.

### Which other providers sell a flat-rate LLM API?

On 15 November 2026, Awan LLM (unlimited tokens with request caps) and Standard Compute (monthly compute budgets) sold flat monthly API plans. Featherless sells a flat Chat plan, but its terms limit it to human-typed chat, with no app or API traffic or background automation.

### Can I try the plan without committing to a month?

Yes. A one-hour Meteor session costs $1 and a one-week Orbit session costs $15, both paid from your balance.

## Keep reading

-   [Cached input tokens: how to cut an LLM API bill](https://broadrouter.com/blog/cached-input-tokens-explained/)
-   [How to run parallel coding agents on a flat-rate LLM API](https://broadrouter.com/blog/run-parallel-coding-agents-flat-rate/)
-   [What "unlimited" really means in an LLM API plan](https://broadrouter.com/blog/what-unlimited-llm-api-really-means/)
-   [A weekly LLM plan for short projects: Orbit vs hour vs month](https://broadrouter.com/blog/weekly-llm-plan-short-projects/)
-   [OpenRouter vs a flat-rate plan for AI agents: which costs less?](https://broadrouter.com/blog/openrouter-alternative-flat-rate/)
-   [LLM API price changes in October 2026: 3 rises, no cuts](https://broadrouter.com/blog/llm-api-price-changes-october-2026/)
-   [Do LLM API prices only go down? 52 rises and how to budget for them](https://broadrouter.com/blog/llm-api-price-increases/)
-   [Guide: How the unlimited LLM API plans work for agents](https://broadrouter.com/guides/unlimited-llm-api-sessions/)
-   [Guide: Why a reasoning model returns an empty answer](https://broadrouter.com/guides/reasoning-tokens-empty-response/)
-   [Guide: Stream chat completions and handle dropped connections](https://broadrouter.com/guides/streaming-chat-completions/)

Try it on your real workload. New accounts start with $5 of credit for Gemma and transcription. DeepSeek, GLM and Qwen3.8 need a top-up.

[Get started](https://inference.broadnet.ai/account/?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-flat-rate-vs-pay-per-token-llm-api&utm_content=page-cta) ·

**Sources**

-   [BroadNet Inference machine-readable prices (pricing.json) (checked 2026-11-15)](https://inference.broadnet.ai/pricing.json?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-flat-rate-vs-pay-per-token-llm-api&utm_content=sources)
-   [BroadNet Inference plans and plan FAQ (checked 2026-11-15)](https://inference.broadnet.ai/plans?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-flat-rate-vs-pay-per-token-llm-api&utm_content=sources)
-   [BroadNet Inference API reference, cached tokens and limits (checked 2026-11-15)](https://inference.broadnet.ai/docs?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-flat-rate-vs-pay-per-token-llm-api&utm_content=sources)
-   [DeepSeek API, models and pricing (fetched 2026-11-15)](https://api-docs.deepseek.com/quick_start/pricing)
-   [Featherless pricing: Chat, Developer and Business plans (checked 2026-11-15)](https://featherless.ai/pricing)
-   [Awan LLM pricing: Core, Plus, Pro and Max plans (checked 2026-11-15)](https://www.awanllm.com/pricing)
-   [Standard Compute pricing: individual and business plans (checked 2026-11-15)](https://standardcompute.com/pricing)
