Claude Opus 5.5 Pricing: API Costs in Dollars and Rupees

The full Opus 5.5 rate card

Fast mode is a research preview on the Claude API only. It runs the same model up to 2.5 times faster at double the standard price, and it isn’t offered on Amazon Bedrock, Google Cloud or Microsoft Foundry.

Claude Opus 5.5 price in India (INR)

Anthropic bills in US dollars. At the mid-market rate of ₹96.5 on 23 September 2026, the main rates work out as follows:

For Indian teams, the cache-read price matters most. Many business assistants resend the same long system prompt, product catalog or policy document on every request. On Opus 5.5 each million of those cached tokens costs about ₹19.3, compared with ₹48.25 on Opus 5.

Compared with Opus 5 and Fable 5.1

Opus 5.5 is 20% cheaper per token than Opus 5 and 60% cheaper than Fable 5.1, while scoring higher than both on Anthropic’s launch benchmarks. The quality side of that trade is covered in Opus 5.5 vs Fable 5.1 and Opus 5.5 vs Opus 5.

What two real workloads cost

These estimates assume each model uses the same number of tokens, so they show the rate-card difference only. In practice Opus 5.5 usually needs fewer, which widens the gap.

Example 1: a customer support assistant

20,000 conversations a month. Each request reads a 4,000-token cached system prompt and help-center extract, adds 600 tokens of new customer text, and produces 1,200 output tokens including thinking at low effort.

Example 2: an agentic coding workflow

200 agent sessions a month. Each session runs about 40 turns over a 60,000-token cached codebase context, which adds up to 2.4 million cached tokens read, 120,000 fresh input tokens and 160,000 output tokens.

In both examples, output dominates the bill. That is why the effort level, which controls how much the model thinks, is the biggest cost lever you have.

What the rate card doesn't show

• Thinking is always on. Opus 5.5 can’t run with thinking disabled, and every thinking token bills at the output rate even when its text isn’t returned.
• The default effort changed. Requests that don’t set effort now run at medium instead of high. At a given level, Opus 5.5 also tends to think more per turn than Opus 5, most of all at xhigh and max.
• Cache misses cost more, relatively. With reads at 0.05x of input, a broken cache prefix is a bigger jump in cost than it was on Opus 5. Changing the top-level effort between requests also invalidates the cache.
• Cloud marketplaces price separately. Amazon Bedrock and Google Cloud set their own Claude prices. Microsoft Foundry bills at Anthropic’s standard rates.

Five ways to cut your Opus 5.5 bill

• Set effort per route. Anthropic reports that on several coding evaluations low effort comes close to medium at much lower cost. Our Opus 5.5 effort levels guide shows how to choose.
• Cache everything stable. Put the system prompt, tool definitions and reference documents first and mark them for caching. The minimum cacheable prompt is 512 tokens.
• Keep conversations append-only. Editing earlier turns breaks the cache and, on Opus 5.5, can invalidate thinking blocks as well.
• Batch anything that can wait. Overnight reports, bulk classification and document processing cost half through the Batch API.
• Skip fast mode unless latency pays for itself. It doubles the price for the same output.

For the full picture of the model behind these prices, read the Claude Opus 5.5 guide.