Tool · Calculator
What does one agent task cost?
Tokens, retries, prompt caching and model routing. Cost per task and per month, with cited prices and the date each was checked.
Cache write is the 5-minute price.
Cache writes cost $2.5/1M here, more than $2/1M input.
Result
Per task
$0.111
₪0.40
Per month
$111.15
₪400
Cache effect per task
$0.071
saved vs. no cache
Pricing sources
- Anthropic https://platform.claude.com/docs/en/about-claude/pricing
Last checked: 2026-10-09 · Cache writes use the 5-minute price. Claude 4.7+ models use a newer tokenizer that produces about 30% more tokens for the same text (per the source page). Token counts you enter are not adjusted. - OpenAI https://developers.openai.com/api/docs/pricing
Last checked: 2026-10-09 · Standard tier, USD per 1M tokens. Short context means up to 272K input tokens, long context means more than 272K (per the page). The page lists a promotional price for gpt-5.6-sol available at least through 2026-11-21; re-check after that date. - Google https://ai.google.dev/gemini-api/docs/pricing
Last checked: 2026-10-09 · Paid tier, standard, USD. Gemini 3.8 Flash $0.75 input / $3.75 output / $0.075 cached input is valid through 2026-12-31; from 2027-01-01 the source lists $1.50 / $7.50 / $0.15. Context caching also has a separate hourly storage fee that is not modelled.
| Model | Input | Cache read | Output |
|---|---|---|---|
| Anthropic · Claude Opus 5.5 | $4 | $0.2 | $20 |
| Anthropic · Claude Sonnet 5.5 | $2 | $0.1 | $10 |
| Anthropic · Claude Sonnet 4.6 (legacy) | $3 | $0.3 | $15 |
| Anthropic · Claude Haiku 4.5 | $1 | $0.1 | $5 |
| OpenAI · gpt-5.6-sol | $4 / $8 (>272K) | $0.4 / $0.8 | $20 / $30 |
| Google · Gemini 3.8 Flash | $0.75 | $0.075 | $3.75 |
Not included: Claude Haiku 5.5 (price depends on prompt length, two tiers) and every OpenAI model except gpt-5.6-sol (only that row was verified against the OpenAI pricing page). Use Custom pricing to enter any other model from the vendor page.
This is an estimate, not an invoice. For gpt-5.6-sol, the context tier changes automatically at 272K input tokens. Other prompt-length tiers, paid tool calls, cache storage fees, volume discounts, or tokenizer differences between model generations are not modeled. Check the source before deciding. All math runs in your browser and nothing you enter is sent anywhere. The site only counts that the calculator was used and that a summary was copied, with fixed labels and no numbers from your inputs.