Anvat / Tools
LLM cost calculator
Estimate your monthly Claude, GPT, Gemini, or DeepSeek bill — with prompt caching, real workload presets, and a side-by-side Anvat discounted projection. Share any calculation via URL.
Start from a typical workload
Default for Claude Code
Tokens sent to the model — system prompt + conversation + tool definitions.
Tokens generated in the response.
Fraction of input tokens served from cache at ~10% of input price (Claude Code typically: 60-85%).
Typical monthly cost at a glance
Common production workloads, costed at both list price and the 30%-off Anvat effective rate. Numbers reflect the same formula the calculator above uses, with realistic cache hit rates baked in.
| Workload | Model | List / mo | Anvat / mo | Savings / yr |
|---|---|---|---|---|
Claude Code · individual Solo developer using Claude Code 4-6 hours/day with caching | Claude Sonnet 4.6 | $352 | $246 | $1.3K |
Claude Code · 10-person team 10 developers, full-day usage, prompt caching enabled | Claude Sonnet 4.6 | $3.5K | $2.5K | $12.7K |
Cursor · power user Heavy autocomplete + Composer usage, no caching | Claude Sonnet 4.6 | $1.4K | $945 | $4.9K |
Coding agent SaaS Production agent backend, 50K active users/mo | Claude Opus 4.8 | $142.2K | $99.5K | $511.9K |
Customer-support chatbot Short turns, broad knowledge needed | GPT-5 mini | $1.4K | $1.0K | $5.2K |
RAG knowledge-base Q&A Doc retrieval + answer synthesis with large cached system prompt | Claude Sonnet 4.6 | $7.8K | $5.4K | $28.0K |
All prices in USD. Numbers exclude Anthropic batch-API discount (additional 50% off on async workloads).
How the math works
- 1. Input tokens split into cached (paid at ~10% of input rate) and fresh (paid at full input rate).
- 2. Output tokens always billed at full output rate.
- 3. Per-request cost = (fresh × input_rate) + (cached × cache_rate) + (output × output_rate), all per million tokens.
- 4. Multiply by requests/day × 30 days for monthly cost.
- 5. Anvat discount of 30% applied to every rate (input, output, cache).
What this calculator doesn't model
- • Batch API discount (50% off, async-only). Stack on top for offline jobs.
- • Anvat 2× credit match on prepaid packs — another 50% effective discount on the topped-up balance.
- • Model routing (Sonnet-default, escalate to Opus on hard tasks) — typically another 40-60% saving.
- • Failed requests, retries, function-call overhead — usually <5% of bill.
See your real bill, not just an estimate
Sign up free, get $2 credit, run a few real requests — Anvat shows the actual discounted rate per request in your dashboard.
Start free →