Flat-rate vs per-token LLM cost calculator
Updated
In short: A flat-rate plan beats paying per token only above a break-even volume: break-even = plan price ÷ your per-token cost per million tokens. For DeepSeek V4.1 Flash with 80% cache hits and 20 input tokens per output token, Full Moon ($60 a month) breaks even near 917M input and 46M output tokens a month.
Estimates, before tax, from BroadNet's published prices synced 2026-11-14: DeepSeek V4.1 Flash at $0.175 input, $0.0055 cached input and $0.52 output per 1M tokens. Every plan runs DeepSeek V4.1 Flash only, one request at a time per session. Pay-as-you-go keys allow 4 requests in flight and 30 a minute. Reasoning tokens count as output.
Report a wrong price or number (email to info@broadnet.ai, with this page's link and your inputs filled in).
How it works
- Enter your monthly volume as input and output tokens in millions, or as requests a month times tokens per request.
- Set the share of input read from cache, how many agents run at the same time, and their active hours per day and days per month.
- Read the comparison: per-token DeepSeek V4.1 Flash against Full Moon, Orbit (weekly price × 30/7) and Meteor (hourly) sessions, with the break-even volume.
- Use the chart to see where the per-token line crosses each flat plan. Copy the link or download CSV to share it.
The formula
per-token / month = ( input × (1 − cache) × input_price
+ input × cache × cached_price
+ output × output_price ) / 1,000,000
sessions = max( agents, ceil(output / (speed × active seconds)) )
Full Moon = sessions × month price
Orbit = sessions × week price × 30 / 7
Meteor = sessions × hour price × active hours × active days
break-even volume = current volume × plan cost / per-token costPer-token cost grows in a straight line with volume; a flat plan is a horizontal line. They cross at the break-even volume, at your current mix of input, output and cache. Doubling agents doubles the flat cost but not the per-token cost, because per-token bills tokens, not parallelism. See flat-rate LLM plan and pay as you go.
Reference workloads
Per-token cost for DeepSeek V4.1 Flash each month, against one Full Moon session at $60.
| Workload | Input / month | Cache | Output / month | Per token | Cheaper |
|---|---|---|---|---|---|
| Side project | 100M | 50% | 10M | $14.23 | Per token |
| Daily assistant | 300M | 50% | 30M | $42.68 | Per token |
| Full-time coding agent | 1,000M | 80% | 50M | $65.40 | Full Moon |
| Always-on agent | 3,000M | 80% | 150M | $196.20 | Full Moon |
Worked example
A coding agent that sends 1,000 million input tokens a month, 80% from cache, and writes 50 million output tokens costs $65.40 per token: $35 for uncached input, $4.40 for cached input and $26 for output. One Full Moon session is $60, or $45 for the first 3 months with a founders code. Three such agents in parallel use three times the tokens and need three sessions, so the ratio stays the same.
Honest caveats
- A session's ceiling is time, not tokens: about 100 output tokens a second, one request at a time. Long prompts also take time to read, so real throughput is lower.
- The plans run DeepSeek V4.1 Flash only. If you need another model, you pay per token for it.
- Measure a week of real
usagedata first. The $5 starting credit and a $1 Meteor hour are cheap ways to test.
Questions
Is an unlimited LLM API plan cheaper than paying per token?
Only above a break-even volume. For DeepSeek V4.1 Flash, with 80% of input read from cache and 20 input tokens per output token, Full Moon ($60 a month) breaks even at about 917 million input and 46 million output tokens a month. Below that, per-token is cheaper.
What does "one session" mean on an unlimited plan?
A session runs one request at a time, like one agent asking, reading the answer and asking again. Agents that work in parallel each need their own session, so three parallel agents cost three sessions. How the sessions work.
How many tokens can one plan session really use in a month?
A session is limited by speed, not a meter. At 100 output tokens per second around the clock, that is about 259 million output tokens in 30 days. Real agents pause between steps, so plan on much less. The calculator warns when your output needs more sessions than agents.
Should I buy a monthly, weekly or hourly plan?
Compare them on the same month. Orbit is $15 a week, about $64.29 over 30 days. Meteor is $1 an hour, so it only wins if you run agents a few hours a month. The calculator prices all three from your active hours.
Do cached tokens change the break-even point?
Yes, a lot. Cached input on DeepSeek V4.1 Flash costs $0.0055 per 1M tokens against $0.175 uncached, so a high cache rate makes per-token cheaper and moves the break-even to higher volume. Cached input tokens explained.
Does a flat plan have rate limits?
The plan caps concurrency: one request in flight per session, with no token meter, no 5-hour window and no weekly quota. A pay-as-you-go key allows 4 requests in flight and 30 a minute. Check your throughput.