Claude Opus API pricing: Opus 5.5 vs Opus 5
Opus 5.5 costs $4 per 1M input and $20 per 1M output tokens; Opus 5 costs $5/$25. A 30-step coding agent task with 8,000 input and 2,000 output tokens per step costs $2.16 on Opus 5.5.
Nearest-priced models, this workload
Showing the 9 models closest in price; change any input to rank every model.
| Model | Per call | Per 1K calls |
|---|---|---|
| Gemini 3.1 Pro Preview | $0.0400 | $40.00 |
| GPT-5.4 | $0.0500 | $50.00 |
| Claude Sonnet 4.6 | $0.0540 | $54.00 |
| GPT-5.6 Sol | $0.0720 | $72.00 |
| Claude Opus 5.5 | $0.0720 | $72.00 |
| GPT-5.5 Rosalind Research | $0.0900 | $90.00 |
| GPT-5.5 Rosalind Discovery | $0.0900 | $90.00 |
| Claude Opus 5 | $0.0900 | $90.00 |
| GPT-5.5 | $0.1000 | $100 |
o4-mini is shut down; its last published list rate is kept for comparing old invoices, not as a model you can call today — the model page names its replacement.
What Claude Opus costs on the Anthropic API
Opus 5.5 is the current Opus tier: $4 per 1M input tokens and $20 per 1M output tokens, with cache reads at $0.2 per 1M. Opus 5, the version it replaced, is $5/$25 with $0.5 cache reads. Both bill every input token, every cached hit and every output token, and reasoning tokens are billed as output.
| Model | Input /1M | Cached read /1M | Output /1M | Checked |
|---|---|---|---|---|
| Claude Opus 5.5 | $4 | $0.2 | $20 | 2026-11-15 |
| Claude Opus 5 | $5 | $0.5 | $25 | 2026-11-15 |
Rates copied from the Anthropic pricing page and checked on the dates shown. Prices change without notice; the date column is how you tell whether a number is current.
Opus 5.5 vs Opus 5: cheaper at every rate
Moving a workload from Opus 5 to Opus 5.5 cuts the input rate by 20% ($5 to $4 per 1M), the output rate by 20% ($25 to $20) and the cached-read rate by 60% ($0.5 to $0.2). On the coding-agent example below that is $2.16 per task instead of $2.70. Keep Opus 5 only if an evaluation set, a pinned prompt or a tool contract was built against it.
What a Claude Opus task actually costs
Standard list rates, no batch or long-context tier. The token counts are stated so you can replace them with your own logs; the calculator above runs the same arithmetic.
| Workload | Opus 5.5 per task or call | Opus 5 per task or call | Volume | Opus 5.5 monthly |
|---|---|---|---|---|
| Coding agent step (repo context in, patch out) 8,000 in / 2,000 out | $2.16 | $2.70 | 30 steps per task, 200 tasks a month | $432 |
| Long-document pass (full report in, summary out) 100,000 in / 3,000 out | $0.4600 | $0.5750 | 500 documents a month | $230 |
| Answer with reasoning tokens billed as output 2,000 in / 4,000 out | $0.0880 | $0.1100 | 20,000 answers a month | $1,760 |
| RAG answer, 80% of the prompt cached 20,000 in / 1,000 out | $0.0392 | $0.0530 | 50,000 answers a month | $1,960 |
Coding-agent and document costs multiply the per-call cost by the number of calls in one task. The monthly column applies your volume to the same token mix. Batch API work is billed at half rate where Anthropic publishes it; check the official page before you budget on it.
When Opus is the right tier
Opus earns its price on multi-step agents where one wrong step wastes the whole run, on long documents where the answer depends on details spread across 100K+ tokens, and on reasoning work you would otherwise run as several Sonnet passes. For high-volume chat, RAG and batch jobs, Claude Sonnet 5.5 at $2/$10 per 1M is half the Opus 5.5 rate on both sides. The rule that holds in practice: price the whole task, not the single call, then compare cost per correct answer.
Three Claude Opus workloads, priced by the token
Opus 5.5 bills $4 per 1M input tokens and $20 per 1M output tokens. A rate card only turns into a budget once a token count sits next to it. The three shapes below cover most Opus work, and every figure comes from the rates above.
A 10,000-token prompt with a 1,000-token answer
One call costs $0.0600 and 1,000 calls cost $60.00. Output is 33% of that bill although it is 9% of the tokens, which is the pattern that repeats on Opus work: the answer is where the money is. The same prompt on Claude Opus 5 costs $0.0750 a call, so the version change alone takes 20% off the request.
A retrieval answer: 5,000 context tokens in, 2,000 out
Retrieval moves more text into the prompt than it writes back. At list rates that answer costs $0.0600, or $60.00 per 1,000 answers. When the same documents are retrieved again and again, put them in a cached prefix: cache reads bill at $0.2 per 1M against $4 for standard input. With 80% of the prompt cached the answer drops to $0.0448, and the prompt caching guide prices the break-even including the write charge.
A 40-turn agent session
Agent cost hides in the turns. Assume a 60,000-token prefix (repo map, tool definitions, instructions) written to cache on turn one and read from cache on the next 39 turns, with 3,000 fresh input and 2,000 output tokens per turn. At the 5-minute cache-write rate of $5 per 1M, the session bills $2.84; 1,000 sessions bill $2,836. Output is 1410437% of that, so a max-output cap saves more than prompt trimming. The same session on Opus 5 costs $4.13 (31% more). If your turns are spaced beyond five minutes, the 1-hour cache at $8 per 1M to write adds $0.1800 per session, which is real money at a few thousand sessions a day.
Opus prices since 4 November 2026
The official Anthropic pricing page has been fetched and diffed on 101 completed checks since our monitoring began on 2026-11-04, and none of those checks flagged a change to the list prices above. The row used here was last confirmed against that page on 2026-11-15. Failed fetches are skipped rather than counted, so the number is checks, not days.
Two levers can move an Opus bill without any list-price change: the 50% Batch API discount Anthropic lists for asynchronous work, and cache writes. The batch discount guide covers the first, the reasoning-token guide covers the second half of the story on reasoning models, where hidden thinking tokens are billed at the output rate and a short visible answer can still be an expensive one.
Opus 5.5 against Sonnet 5.5, Haiku 5.5 and GPT-5.5
Claude Sonnet 5.5 lists $2/$10, exactly half the Opus rate on both sides. Claude Haiku 5.5 lists $0.1/$0.5, which is 40x cheaper on input and 40x on output. Against OpenAI, GPT-5.5 lists $5/$30, so Opus 5.5 is the cheaper of the two premium models per token (20% lower input, 33% lower output). On the 10,000/1,000 request that is $0.0600 on Opus 5.5 against $0.0300 on Sonnet 5.5, $0.0015 on Haiku 5.5 and $0.0800 on GPT-5.5.
The ratio sets the test: at twice the Sonnet rate, Opus 5.5 has to finish a task in fewer than two Sonnet attempts, or remove a review pass that costs more than the difference. On the 40-turn session above that is $1.42 on Sonnet 5.5 against $2.84 on Opus 5.5, so one avoided retry in three roughly pays the gap. Model-level head-to-heads: GPT-6.1 Sol vs Claude Opus 5.5, Claude Opus 5.5 vs Gemini 4 Argon and Claude Fable 5.1 vs Claude Opus 5.5. To turn any of this into a per-user price, the AI SaaS pricing guide and the unit economics calculator do the margin arithmetic; Claude cost calculator runs these token mixes on your own numbers.
Related Claude prices
- Claude Sonnet pricing (Sonnet 5 and 5.5)
- Claude Haiku pricing (the cheap tier)
- Claude Fable 5.1 pricing
- All Claude models on the Anthropic API
- Claude cost calculator (all models, your token mix)
Questions
How much does Claude Opus cost per 1M tokens?
Opus 5.5 is $4 input and $20 output per 1M tokens; cache reads are $0.2 per 1M. Opus 5 is $5/$25 with $0.5 cache reads. Checking dates are on the table above.
Is claude opus pricing the same as claude opus api cost?
The pricing is the rate card; the cost is what your token mix turns it into. The rate card is $4 in and $20 out per 1M on Opus 5.5. The bill is those rates times your tokens, and output plus reasoning tokens usually dominate it.
Is Opus 5.5 cheaper than Opus 5?
Yes. Opus 5.5 is 20% cheaper on both input and output, and its cached-read rate is less than half of the Opus 5 rate card. The only reason to stay on Opus 5 is a pinned evaluation or prompt.
Is Claude Opus worth it over Sonnet?
Price a task, not a call. Opus 5.5 is twice the Sonnet 5.5 rate on both sides ($4/$20 against $2/$10), so Opus wins only when it removes retries or review work that costs more than the difference.
Do Opus prices include batch or long-context discounts?
No. The rates above are standard short-context list prices. Anthropic publishes a 50% Batch API discount for models where the official page lists it; long-context pricing is only listed for models that publish a long-context tier. The official page linked above is the only source to budget from.
Next step → Estimate cost per user with Claude Opus
Price data last checked 2026-11-18 against official sources · how we verify
Price-change alerts
One short email when a provider changes a price or a promo ends. No spam, unsubscribe any time.