Anthropic sells Claude in three shapes: Claude Pro ($20/mo), Claude Max ($200/mo), and pay-as-you-go API. They're priced for completely different users, and most developers pick the wrong one — usually the most expensive. This article runs the actual numbers across three realistic workloads so you can see, in dollars per month, which option fits your work.
Tiers at a glance
| Tier | Price | Limits | Best for | Worst for |
|---|---|---|---|---|
| Claude Pro | $20/mo | ~45 msgs / 5h, 5× Free | Chat, research, writing | Heavy Opus coding |
| Claude Max | $200/mo | 5–20× Pro, priority on Opus 4.7 | Heavy daily chat-based coding | Irregular use |
| Claude API | pay-as-you-go | $3/M input, $15/M output (Sonnet) | Dev, scripts, agents, production | One-off chats |
| API + Claudexia | same per-token | + SBP/RU card, key in 60s | Any API workflow from blocked region | When you already have a working US card |
What each tier actually buys
Claude Pro — $20/month. Browser and desktop chat with Claude. Soft limits of roughly 45 Sonnet messages per 5-hour window and 5 Opus messages. Good if you use Claude as a chat companion: copy review, planning conversations, document drafts. Not designed for IDE work or agents — the rate limits make sustained development impossible.
Claude Max — $200/month. Same shape as Pro with limits multiplied by 5-20x, plus priority Claude Code access in the browser and longer Opus sessions. Works for a developer who lives in Claude Code 6+ hours a day. Anything less and you're overpaying.
Claude API — pay-as-you-go. Per-million-token pricing per model, no message limits, no 5-hour windows. Your scripts and agents run 24/7 if they need to. You pay exactly what you consume.
Three real workloads, real money
Take a typical backend developer working in Python/TypeScript using Claude Code daily.
Workload 1: Light usage
10-15 PRs per week, mostly small refactors and bug fixes. Sonnet 4.6 as default, Opus on ~10% of tasks.
- Monthly tokens: ~5M input (with caching, effective ~1.5M), ~600K output.
- API cost: Sonnet 4.6 ~$4-6, Opus 4.7 ~$1-2. Total ~$7/month.
- Claude Pro ($20): Opus rate-limit hits 4-5 times a week.
- Claude Max ($200): 28x overpayment.
Workload 2: Active development
30-50 PRs per week, agent pipelines, heavy Sonnet usage, regular large Opus refactors.
- Monthly tokens: ~25M input (with caching ~6M), ~3M output.
- API cost: Sonnet 4.6 ~$25, Opus 4.7 ~$15-20. Total $40-50/month.
- Claude Pro ($20): hits Sonnet limits daily.
- Claude Max ($200): unlimited but 4-5x overpayment.
Workload 3: 5-developer team
Everyone on Sonnet 4.6 + Opus 4.7, ~30 PRs per week each.
- API per developer: ~$40-50/month.
- Claude Max ×5: $1000/month.
- API through Claudexia: ~$250/month for the whole team (with prompt caching and a shared balance).
Annual savings vs Max ×5: $9,000/year — roughly an extra junior engineer.
Why API on Claudexia is cheaper still
Claudexia forwards cache_control headers transparently. Claude Code and any agent with prompt caching support caches the system prompt, instructions, and part of the context for an hour — re-reading those tokens runs at 0.10x the standard rate.
In real backend workloads, prompt caching cuts the input bill by 60-80%. Without it, Sonnet 4.6 in the active workload would cost ~$80/month; with it, $25. Pro and Max give you no control over caching — it's a black box with a fixed allotment.
When the subscription actually wins
Claude Pro makes sense if:
- You mostly use Claude in the browser for conversation, reviewing documents, copywriting — no Claude Code, no agents.
- You need synchronous chat with Projects and long persistent sessions.
- Your monthly API spend on chat would exceed $25-30 (rare — chat usually costs $3-7/month on API).
Claude Max makes sense if:
- You spend 6+ hours/day in Claude Code and regularly hit Opus rate limits.
- You need priority access during peak Anthropic load.
- You want a fixed monthly bill with zero variance.
Outside those cases, API is cheaper — often by a wide margin.
Buying Claude API without an Anthropic account
Claudexia replaces the whole signup-card-billing flow with one API key:
- Sign up at claudexia.tech — email only, no card required.
- Top up with a card, SEPA, SBP, e-wallet, or crypto (USDT/BTC/ETH).
- Create a key (
sk_cdx_…) with per-model and per-day spending caps. - Wire into Claude Code, Cursor, or your scripts via
ANTHROPIC_BASE_URL=https://api.claudexia.tech.
No Anthropic account, no VPN, no virtual card games. No minimum top-up.
FAQ
Is Claude Pro worth it as a backup? Only if you specifically want browser chat with Projects. For everything else, API delivers more usage per dollar.
Pro vs ChatGPT Plus? Both ~$20/month for chat. For development work, API beats both by a wide margin — usually 3-10x cheaper for the same throughput.
How do I forecast token cost before paying? The Claudexia dashboard ships a model-by-model calculator and per-key history. Every Anthropic SDK response also includes a usage block — log it and project from there.
Does the balance expire? No. Claudexia balances never expire — they sit there until you spend them.
Can I mix? Yes. Many teams keep one Claude Pro seat for browser chat and put everyone else on a shared Claudexia API key with per-developer spend caps.