OmniscioToolsClaude Code cost calculator
Claude Code cost calculator
Work out what a fleet of Claude Code agents costs per hour, per day and per month at API prices. Choose a model, set how many agents run at once and for how long, and the figures update as you type. Every price and assumption is listed below the calculator.
Prices checked October 12, 2026. Updated .
Estimate your fleet
The same fleet on every model
The workload you set above, priced on each model. Switching the model is the biggest single lever on the bill, and parallel agents make it easy to do per task: a cheap model for the mechanical work and a frontier model for planning and hard debugging.
| Model | Per agent-hour | Fleet per month |
|---|---|---|
| Claude Fable 5.1 | $10.08 | $4,233.60 |
| Claude Opus 5.5 | $4.46 | $1,874.88 |
| Claude Sonnet 5.5 | $2.23 | $937.44 |
| Claude Haiku 5.5 | $0.13 | $55.94 |
| DeepSeek V4.1 Flash | $0.24 | $101.61 |
How the estimate works
Claude Code talks to the model in requests. Every request carries the conversation so far, and every tool call (reading a file, running a command, editing code) ends one request and starts the next. So the cost of an hour is the number of requests multiplied by what each one costs.
Each request is priced in three parts. The share of the context already in the prompt cache is billed at the cache-read rate. The rest is fresh input, which Claude Code writes to the cache as it goes, so on Claude models it is billed at the five-minute cache-write rate (1.25 times the input price); DeepSeek has no cache-write charge and bills it as ordinary input. The tokens the model writes back, thinking included, are billed at the output rate.
A day is that hourly cost multiplied by the active hours you set, and the month multiplies it by the working days and the number of agents. On Claude Haiku 5.5 a request whose context is over 100,000 tokens is billed at its higher long-prompt prices, and the calculator switches to them on its own.
The prices it uses
US dollars per million tokens, from Anthropic's pricing page and DeepSeek's. DeepSeek's figures are its peak-hour rates; outside its peak hours it bills half.
| Model | Input | Cache write | Cache read | Output |
|---|---|---|---|---|
| Claude Fable 5.1 | $10.00 | $12.50 | $0.25 | $50.00 |
| Claude Opus 5.5 | $4.00 | $5.00 | $0.20 | $20.00 |
| Claude Sonnet 5.5 | $2.00 | $2.50 | $0.10 | $10.00 |
| Claude Haiku 5.5 (prompts up to 100,000 tokens) | $0.10 | $0.125 | $0.01 | $0.50 |
| DeepSeek V4.1 Flash | $0.30 | $0.30 | $0.006 | $1.20 |
Over 100,000 tokens of prompt, Claude Haiku 5.5 bills $0.50 input, $0.625 cache write, $0.05 cache read and $2.50 output. This is an estimate at list prices: batch discounts, US-only inference (1.1 times) and negotiated rates are left out.
Where the default assumptions come from
The defaults (60 requests an hour, 80,000 tokens of context, 90% read from cache, 1,000 tokens out) describe a steady coding session. On Sonnet 5.5 they come to about $9 for a four-hour day, below Anthropic's reported average of about $13 per active day. Your own numbers are better: Claude Code's /usage command shows a session's input, output, cache-read and cache-write tokens, and dividing by the hours it ran gives you the inputs for this page.
What moves the number
- The model. On the default workload, Opus 5.5 costs twice what Sonnet 5.5 does and Fable 5.1 about four and a half times. Reserve the expensive model for the work that needs it.
- Context size. Every request re-sends the conversation, so a session left open across unrelated tasks pays for all of them on each turn.
/clearbetween tasks resets it. - Cache misses. On an API key the cache lasts five minutes by default. An agent that sits idle past that pays to re-send its whole context at the write rate on the next request. Drop the cache share above to 50% to see what that does.
- Agents that never stop. An agent retrying the same failing command keeps billing every request. Anthropic's advice for agent teams applies to any fleet: shut agents down when their work is done.
Where Omniscio fits
Omniscio is the desktop app we make for running many Claude Code sessions at once. Three parts of it are about the bill this page estimates.
- Each session shows its running cost in its header, so the estimate on this page becomes a real figure per session.
- A session on a pay-per-use key can carry a per-session hard cap: when its cost crosses the limit you set, Omniscio stops it, with a warning before that. Subscription sessions are never capped.
- Each session can run on its own provider, so a mechanical task can use a cheaper model than the one doing the planning. DeepSeek, Kimi, GLM and MiniMax are in the list alongside Claude; switch on the ones you use in Settings and add the key or sign-in there.
It runs on Windows, macOS and Linux. The features page lists everything in it today.
Common questions
How much does Claude Code cost per day?
Anthropic's own figure, across enterprise deployments, is about $13 per developer per active day and $150 to $250 per developer per month, with 90% of users under $30 per active day. Running several agents at once multiplies that roughly by the number of agents.
Does running agents in parallel cost more per agent?
No. Each agent has its own context window and is billed for its own requests, so the bill grows in line with the number of agents and the hours they run. Five agents for four hours costs about the same as one agent for twenty.
Why does the prompt cache matter so much?
Claude Code sends the whole conversation with every request. A cache hit on Sonnet 5.5 costs $0.10 per million tokens against $2 for fresh input, so a session that keeps its cache warm pays a fraction of one that keeps missing it.
Does this apply to a Claude subscription?
No. This calculator prices pay-as-you-go API usage. A Pro or Max subscription is a flat monthly fee with usage limits, so it is billed differently; the estimate is still a useful measure of how much work a fleet is doing.
See the real cost of every session
Install Omniscio, start your agents, and watch each session's cost as it runs. The guide to running multiple sessions covers the rest of the setup.