Anthropic
Claude Sonnet cost & gross margin per customer
Claude Sonnet is Anthropic's balanced model, popular for coding and long-context work. Its output tokens are pricier than GPT-4o's, so output-heavy use cases need a closer eye on per-customer margin.
Claude Sonnet's output at $15/Mtok is the priciest of the mainstream mid-tier models, so output-heavy use — long completions, code generation — is where your margin lives or dies. Input is reasonable, which helps for long-context, retrieval-style workloads.
Input
$3 /Mtok
Output
$15 /Mtok
Margin per customer by usage & plan price
How Claude Sonnet margin holds up as a customer's usage rises, across common subscription prices.
| Usage / mo | LLM cost | $19/mo | $29/mo | $49/mo | $79/mo |
|---|---|---|---|---|---|
| Light | $0.75 | 96.1% | 97.4% | 98.5% | 99.1% |
| Typical | $3.75 | 80.3% | 87.1% | 92.3% | 95.3% |
| Heavy | $15.00 | 21.1% | 48.3% | 69.4% | 81% |
| Power user | $61.50 | -223.7% | -112.1% | -25.5% | 22.2% |
Margin % per customer at each plan price. Token prices indicative, as of 2026-06.
Of the $3.75 a typical customer costs on Claude Sonnet, output tokens are $2.25 (60%) and input $1.50. Output is priced at $15/Mtok — 5× the input rate — so the more your product generates per request, the faster a customer's margin slips.
Worked example
Take a power user on your $49/mo plan sending 8M input / 2.5M output tokens a month. On Claude Sonnet that's $61.50 in tokens — you're losing $12.50 on that single customer every month (-25.5% margin).
How to keep Claude Sonnet profitable
- Trim and cache input context — long system prompts and re-sent chat history are pure, repeated cost.
- Cap output length and stop generation early where you can: at roughly 5× the input price, every extra generated token is where Claude Sonnet hurts most.
- Route easy requests to a cheaper model and reserve Claude Sonnet for the hard ones that actually need it.
- Set a per-customer margin alert so one heavy user can't quietly slip into the red unnoticed.
When to choose Claude Sonnet
Choose Claude Sonnet for coding, long-context, and quality-sensitive work — just price output-heavy plans accordingly. It rewards products that charge enough to cover generous generation.
FAQ
- How much does Claude Sonnet cost per customer?
- At a typical 500k input / 150k output tokens per customer per month, Claude Sonnet costs about $3.75 per customer (input 3/Mtok, output 15/Mtok).
- Is Claude Sonnet profitable for a $49/mo AI SaaS?
- At typical usage, yes — margin is about 92.3% ($45.25 per customer). It erodes as usage rises; heavy and power users are where Claude Sonnet can turn unprofitable.
- What's a good gross margin for an AI SaaS using Claude Sonnet?
- Most AI products target a 60–80% gross margin. With Claude Sonnet at typical usage you're around 92.3% on a $49 plan — comfortable — but your blended margin depends on the heavy users, which is the number worth watching.
- At what usage does Claude Sonnet stop being profitable on a $29 plan?
- Around 3.9M input / 1.2M output tokens a month. Past that point, a $29 customer costs you more than they pay.
- How do I reduce Claude Sonnet cost per customer?
- Cut output tokens first (they're the priciest), cache or trim input context, route easy requests to a cheaper model, and watch the break-even point — around 6.5M input / 2.0M output tokens a $49 customer stops being profitable.
Compare this model
Other models
Key terms