Claude API pricing for business chatbots: what a conversation really costs
As on 17 September 2026, Anthropic lists Claude Haiku 4.5 at $1 input and $5 output per million tokens, Claude Sonnet 5 at $2 and $10, and Claude Opus 5 at $5 and $25. Cached prompt reads cost a tenth of the input price, and the Batch API is 50% off. On Convo360 you add your own Claude key and pay Anthropic directly, with zero markup.
Prices as on 17 September 2026. Figures were checked against the provider's official pages on that date. Providers change prices and rules without notice, so confirm on the official page before you budget. Convo360 adds no markup: you pay the provider directly with your own account. Convo360 is not affiliated with Anthropic. Claude and its model names are trademarks of their owner. Prices were read from Anthropic's official pages on 17 September 2026 and may change; always confirm on the official pricing page.
- Claude is billed per million tokens in US dollars, with output tokens priced at five times input tokens on every current model.
- As on 17 September 2026 the published input and output rates are $1 and $5 for Haiku 4.5, $2 and $10 for Sonnet 5, and $5 and $25 for Opus 5.
- Prompt caching charges 1.25 times the input price to write a 5-minute cache and 0.1 times to read it, so a long repeated prompt gets much cheaper.
- API usage is prepaid: Anthropic says credits expire one year after purchase and are non-refundable, so top up in small amounts.
- Convo360 is BYOK: your Claude key goes in Settings → Provider Keys, you pick the model per agent, and Anthropic bills you directly.
Claude API prices as on 17 September 2026
These figures were read from Anthropic's official pricing pages on 17 September 2026. All prices are in US dollars per million tokens (MTok). The table covers the models a business chatbot is likely to use; the complete list, including legacy models, is on Anthropic's pricing page.
| Model | Input | Output | 5-minute cache write | Cache read |
|---|---|---|---|---|
| Claude Haiku 4.5 | $1 | $5 | $1.25 | $0.10 |
| Claude Sonnet 5 | $2 | $10 | $2.50 | $0.20 |
| Claude Opus 5 | $5 | $25 | $6.25 | $0.50 |
| Claude Fable 5.1 | $10 | $50 | $12.50 | $0.25 |
| Claude Sonnet 4.6 (listed under legacy models) | $3 | $15 | $3.75 | $0.30 |
Anthropic describes Haiku 4.5 as its fastest, most cost-efficient model. Its docs note that the $2 and $10 Sonnet 5 rate began as introductory pricing and is now the standard price. Older models such as Claude Haiku 3.5 are marked retired on Anthropic's own API.
Convo360 is not affiliated with Anthropic. Prices can change at any time, and Anthropic's page is the final word.
What you are billed for: tokens, credits and tax
Every request is charged on two counts: the tokens you send (system prompt, knowledge base text, chat history and the customer's message) and the tokens Claude writes back. Anthropic's rule of thumb is one token for roughly 4 characters or 0.75 words in English, varying by language. Hindi and Hinglish tokenise differently, so measure on real chats.
- Tokenizer: Anthropic's docs say Claude 4.7 and later models use a newer tokenizer that produces approximately 30% more tokens for the same text. Sonnet 5 and Opus 5 fall in that group; Haiku 4.5 uses the previous one.
- Long context: Claude 4.6 and later models include the full 1M token context window at standard pricing.
- Prepaid credits: API usage is paid from credits bought in the Claude Console. Auto-reload can top up when the balance falls below a threshold you set.
- Expiry: Anthropic states that credits expire one year from purchase and are non-refundable.
- Failed calls: failed requests are not charged. When credits run out, the API stops until you add more.
- Free credits: the docs say new users receive a small amount of free credits for testing, without stating the amount.
The pricing page carries the note that prices shown do not include applicable tax. It does not explain how Indian GST applies to API credits, so check your first invoice.
How prompt caching changes the cost of a long prompt
A business bot resends the same system prompt and knowledge base text on every turn. Prompt caching stores that repeated prefix. Anthropic prices it with multipliers on the input rate: a 5-minute cache write costs 1.25 times, a 1-hour cache write costs 2 times, and a cache read costs 0.1 times (0.025 times on Claude Fable 5.1). The 5-minute cache is refreshed at no extra cost each time it is used, and Anthropic says it pays off after a single cache read.
ILLUSTRATIVE example on Claude Sonnet 5: a 5,000-token system prompt plus knowledge base, sent 10 times within a few minutes.
| Approach | Calculation | Cost of the repeated part |
|---|---|---|
| No caching | 10 × 5,000 tokens × $2 per MTok | $0.10 |
| With caching | 1 write at $2.50 per MTok, then 9 reads at $0.20 per MTok | about $0.0215 |
Three conditions apply. The prompt must cross a minimum length: 1,024 tokens on Sonnet 5, 4,096 on Haiku 4.5 and 512 on Opus 5, per Anthropic's caching docs. The cached text must be identical each time. And caching has to be requested by the software that calls the API, so it is not automatic. Open the usage report in your Claude Console: if cache read tokens show zero, budget at the normal input price.
A per-conversation cost formula
The numbers below are ILLUSTRATIVE. Replace them with token counts from your own Console usage report.
Formula: cost per conversation = (total input tokens ÷ 1,000,000 × input price) + (total output tokens ÷ 1,000,000 × output price).
Illustrative chat: six bot replies. Each turn resends the prompt and the growing history, so input adds up to 18,000 tokens across the chat. The replies total 900 output tokens.
| Model | Input cost | Output cost | Per conversation | Per 1,000 conversations |
|---|---|---|---|---|
| Claude Haiku 4.5 | $0.018 | $0.0045 | $0.0225 | $22.50 |
| Claude Sonnet 5 | $0.036 | $0.009 | $0.045 | $45 |
| Claude Opus 5 | $0.09 | $0.0225 | $0.1125 | $112.50 |
The token count is kept equal to show the rate difference only; the newer tokenizer counts the same text as more tokens on Sonnet 5 and Opus 5. Anthropic's own docs give a support example of about 3,700 tokens per conversation on Haiku 4.5, working out to roughly $37 per 10,000 tickets. For the same exercise on another provider, see estimating OpenAI API cost for a chatbot.
Paying for Claude from India
Anthropic states that all prices and payments are in USD, with credit cards accepted for standard accounts. We do not quote a rupee price because the rate changes daily. Use the formula instead:
₹ cost = USD amount × your bank's rate on the day + GST and forex fees, if any.
Since the charge is in USD, keep international transactions enabled on your card. And because credits expire after one year and cannot be refunded, buy what you expect to use in a few months, then set auto-reload once volume is steady.
How to add your Claude key in Convo360
Convo360 is BYOK. You bring your own Anthropic account, Anthropic bills you directly, and Convo360 adds zero markup on tokens. Plans start at ₹999 per month, excluding 18% GST.
- In the Claude Console, create an API key and buy a small amount of credits.
- In Convo360, open Settings → Provider Keys, choose Anthropic (Claude), paste the key and click Test connection. Keys are encrypted at rest.
- Open AI Agents, edit your agent, and pick a Claude model from the model dropdown under the brain and knowledge settings. Match the name in the dropdown to the row in Anthropic's table to know your rate.
- Attach a knowledge base, then use Preview / Try-it with real customer questions before publishing.
In Convo360, Claude works as the brain of chat agents: WhatsApp, website chat, Instagram and the AI Reply block in flows. Live voice calls use a speech engine that runs on your OpenAI or Gemini key. Details are in the AI agents docs.
If the key is revoked or the credits run out, Anthropic stops answering. Convo360's Provider Health monitoring detects the provider error and shows a clear alert, so you know the fix is a top-up or a new key, not a broken flow.
Ways to cut the Claude bill
- Match the model to the job. Anthropic's own guidance is Haiku for simple tasks, Sonnet for most production workloads and Opus for the most complex reasoning. See how to pick a brain.
- Keep replies short. Output costs five times input on every current model. Ask for two or three sentences in the system prompt.
- Trim the prompt. Every token in the system prompt is paid for again on every turn. Let the knowledge base retrieve the relevant passage instead of pasting whole documents.
- Use flow blocks for fixed steps. Buttons, List Menu and Ask a Question blocks collect a name or a slot without spending a token.
- Keep Batch for offline work. The Batch API is 50% off input and output, but it is asynchronous, so it suits bulk tagging or summaries, not live chat.
Example conversation
Connect your Claude key and test a bot free
Start free in minutes — no code, no credit card. Bring your own AI keys and pay no per-message markup.
Frequently asked questions
Is there a free tier on the Claude API?
Anthropic does not list a permanent free tier for the API. Its docs say new users receive a small amount of free credits to test the API, without publishing the amount. After that, usage is paid from prepaid credits bought in the Claude Console.
Is GST extra on Claude API prices?
Anthropic's pricing page, as read on 17 September 2026, says prices shown do not include applicable tax. It does not explain how Indian GST is applied to API credits, so check your invoice from Anthropic or ask your accountant before you fix a budget.
What is the Claude API price in rupees?
Anthropic publishes and charges in US dollars only. To estimate rupees, multiply the USD amount by your bank's exchange rate on the day and add any forex fee or tax your bank or Anthropic applies. We do not quote a converted figure because the rate changes daily.
Does prompt caching work automatically on the Claude API?
No. Anthropic's docs say the calling application enables caching with a cache control field in the request, and the prompt must cross a minimum length such as 1,024 tokens on Sonnet 5. Your Console usage report shows whether any cache read tokens were billed.
Do Claude API credits expire?
Yes. Anthropic's help centre states that credits expire one year from the purchase date, that the date cannot be extended, and that all credit purchases are non-refundable. Buying smaller amounts more often, or using auto-reload, avoids paying for credits you never use.
Can I use my own Anthropic account with Convo360?
Yes, that is how it works. Convo360 is BYOK, so you paste your own Claude API key under Settings → Provider Keys, choose a Claude model on the AI agent, and Anthropic bills your account directly with zero markup from Convo360.
Official sources
- Anthropic: Claude pricing page, API section (official)
- Claude Platform docs: Pricing (official)
- Claude Platform docs: Prompt caching (official)
- Claude Help Center: How do I pay for my Claude API usage? (official)
Related reading
Available across India — see WhatsApp & AI calling automation by state and city.