What is Opus 5?
Claude Opus 5 is the Opus model Anthropic released on 24 July 2026, between Opus 4.8 and Opus 5.5. Anthropic's launch post calls it "a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price." That price is $5 per million input tokens and $25 per million output, the same as Opus 4.8.

| Claude Opus 5, read 15 November 2026 | Value |
|---|---|
| Released | 24 July 2026 |
| API model id | claude-opus-5 |
| Price per million tokens, input and output | $5 and $25 |
| Cache read, 5 minute cache write, 1 hour cache write | $0.50, $6.25 and $10 per million |
| Batch API | 50% off input and output |
| Fast mode (research preview) | $10 and $50 per million, up to 2.5x output speed |
| Context window and max reply | 1M tokens and 128K tokens |
| Reliable knowledge cutoff | May 2026 |
| Status | Legacy, retirement not sooner than 24 July 2027 |
Every row is from Anthropic's Opus 5 model page and pricing page. It runs on the Claude API and on the cloud platforms that Anthropic lists, and the id is the same on most of them. Thinking is adaptive, and the default effort is high.
The scores are from Anthropic's Opus 5.5 launch post, which set the two models side by side on 22 September 2026. The point gaps are our subtraction. They matter because Opus 5 was the newest Opus for only 60 days, and that is a short life for a flagship.
Is Opus 5 still available?
Yes, but it is no longer the current Opus. Anthropic lists Claude Opus 5.5 as the current Opus and files Opus 5 under legacy. The Opus 5 page says: "Although Claude Opus 5 is still available, you should consider migrating to Claude Opus 5.5." Legacy doesn't mean gone. Anthropic promises that it won't retire Opus 5 before 24 July 2027, so the id keeps working until then. The same page lists claude-opus-5 on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS.
Opus 5 vs Opus 5.5
Opus 5.5 is the cheaper model, and it is the faster one too. Anthropic's launch post says it "performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5." It puts the price at $4 and $20 per million tokens, which is 20% below Opus 5, and says that output comes more than 30% faster. The table sets the two side by side.
| As of November 2026 | Claude Opus 5 | Claude Opus 5.5 |
|---|---|---|
| Released | 24 July 2026 | 22 September 2026 |
| Anthropic status | Legacy | Current |
| Input and output per million tokens | $5 and $25 | $4 and $20 |
| Cache read per million tokens | $0.50 | $0.20 |
| Fast mode, input and output | $10 and $50 | $8 and $40 |
| Default effort | high | medium |
| Reliable knowledge cutoff | May 2026 | June 2026 |
| Terminal-Bench 4.0 | 52.3% | 66.4% |
| GDPval-AA | 1708 Elo | 1846 Elo |
| Earliest retirement | 24 July 2027 | 22 September 2027 |
Both models share a 1M token window and 128K tokens of output, so capacity isn't the question. Opus 5 costs more per token than its successor and scores lower on Anthropic's own tests. Keep it only if you pinned the id for a reason, such as a prompt that you tuned on it or a regression test that needs the same weights.
Anthropic's models overview now tells developers to "start with Claude Opus 5.5 for most workloads." It lists Fable 5.2 above it at $8 and $40. Claude Sonnet vs Opus covers where Opus 5.5 sits against Sonnet.
Opus 5 price per million tokens
Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens on the API, as of November 2026. A chat answer of 1,000 tokens in and 500 out costs $0.0175. That's our sum from the list price. The same answer on Opus 5.5 costs $0.014, which is 20% less.
Three discounts and one surcharge change the bill. Cache reads cost $0.50 per million, which is a tenth of the input price, once a prompt of at least 512 tokens has been cached. The Batch API takes 50% off both rates. Fast mode doubles them to $10 and $50. Anthropic still marks fast mode on Opus 5 as a research preview, and it isn't available with batch. The Claude API pricing guide works through all of them, and the cheapest way to use Claude Opus covers the plan route.
On a subscription you don't pay per token. Claude Pro is $20 a month, or $17 a month billed yearly at $200, per claude.com/pricing, read 15 November 2026. Opus sits on Pro and Max. Free has Sonnet and Haiku only.
Which Claude plans include Claude Opus 5?
Claude Pro and Max include Opus, and Claude Free doesn't. At launch Anthropic said that Opus 5 was "the new default model on Claude Max, and the strongest model on Claude Pro." The plan page, read 15 November 2026, lists Opus as a Yes for Pro and Max and a No for Free. It names no version, and the newest Opus that you will get there is now 5.5.
| Claude plan, as of November 2026 | Price | Opus | Fable |
|---|---|---|---|
| Free | $0 | No | No |
| Pro | $20 a month, or $17 billed yearly | Yes | Paid with usage credits |
| Max | From $100 a month | Yes | Up to half the weekly limit |
| Claude API | Per token | Opus 5 and Opus 5.5 | Fable 5.2 and earlier |
The launch post names only Max and Pro among the consumer plans. It does say fast mode reaches Claude Code through usage credits. It also says flagged cybersecurity requests fall back to Opus 4.8 by default in Claude.ai, Claude Code and Claude Cowork.
How good is Claude Opus 5?
Claude Opus 5 was the top scorer on ARC-AGI-3 at launch: 30.2% at high effort on the ARC Prize leaderboard, as of 24 July 2026. ARC Prize also lists 88.3% on ARC-AGI-2 at high effort and 90.4% at max, and 97.5% on ARC-AGI-1 at both. Anthropic says its ARC-AGI-3 score is about three times the next best model's.
Anthropic's own claims are mostly relative. It says that Opus 5 more than doubles Opus 4.8 on Frontier-Bench v0.1. On CursorBench 3.2 at max effort it lands within 0.5% of the best score of Fable 5, at half the cost per task. On OSWorld 2.0 it beats the best result of Fable 5 for just over a third of the cost. The launch post prints no absolute figure for these, and the text gives none for the charts on GDPval-AA v2, HLE and DeepSearchQA.
Anthropic reports an overall misaligned behavior score of 2.3 in its automated audit, which it says is the lowest of its recent models, and it adds that the model gained 10.2 points on an internal organic chemistry test and 7.7 on protein tasks over Opus 4.8, though all of those are the lab's own tests. Treat them as claims. Wait for someone else to repeat them.
It isn't better at everything. The Batch reported that Opus 5 hallucinates more often and recalls facts less reliably than Fable 5. If you rely on the model alone to get a fact right, test that on your own questions first.
Claude Opus 5 vs Fable 5 and Opus 4.8
Opus 5 costs half of what Fable 5 does, and the same as Opus 4.8. Anthropic prices Fable 5 at $10 and $50 per million tokens and Opus 4.8 at $5 and $25. All three share a 1M token window and 128K tokens of output. Fable has since moved on to 5.2, at $8 and $40.
| Per Anthropic, November 2026 | Input and output per million tokens | Status |
|---|---|---|
| Claude Fable 5.2 | $8 and $40 | Current |
| Claude Opus 5.5 | $4 and $20 | Current |
| Claude Opus 5 | $5 and $25 | Legacy |
| Claude Fable 5 | $10 and $50 | Listed, older |
| Claude Opus 4.8 | $5 and $25 | Listed, older |
Pick Opus 5 only if your code already names claude-opus-5. Pick Opus 5.5 for new work, since it costs less and Anthropic points all developers to it. Pick Fable 5.2 when Opus at higher effort still fails your own tests, which is the advice that Anthropic gives. The smartest AI model page tracks the top scorers across labs.
Can you use Claude Opus 5 without a Claude plan?
You can use Claude Opus 5 on Whizi Pro, which carries Claude, GPT and Gemini in one app. As of November 2026 Whizi lists both Claude Opus 5 and Claude Opus 5.5 on Pro, at 20 credits for each message. Pro includes 2,500 credits a month. The Pro plan costs $29.99 a month, and its first week is $1.99.
That's a different bargain from Claude Pro. Claude Pro at $20 gives you Opus across Anthropic's apps under Anthropic's usage limits, while Whizi Pro gives you a credit pool that you spend on any model, so one message to Opus costs what two to Sonnet do. It suits you if you also use GPT or Gemini and want one login. It won't suit you if you live in Claude Code or call Anthropic's API, because Whizi runs neither. The Claude in Whizi docs list every Claude row.
The free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5, with no card. It doesn't include Opus.
- Opus 5 is $5 and $25 per million tokens with a 1M window, and Anthropic calls it legacy
- For new work use Opus 5.5 at $4 and $20, which scores higher on Anthropic's tests
- Keep
claude-opus-5only for pinned prompts, since it stays up until at least 24 July 2027 - Claude Free has no Opus, so you need Pro, Max, the API or a multi-model app
- Treat vendor scores as claims and test on your own work
Frequently asked questions
When was Claude Opus 5 released?
Anthropic released Claude Opus 5 on 24 July 2026, according to its model page and launch post. Opus 5.5 followed on 22 September 2026. The model id is claude-opus-5.
Is Claude Opus 5 free?
No. Claude Free includes Sonnet and Haiku, but not Opus, as of 15 November 2026 on claude.com/pricing. Opus starts on Claude Pro at $20 a month. On the API it costs $5 and $25 per million tokens.
What is the Claude Opus 5 context window?
The context window is 1M tokens and the max reply is 128K tokens on the synchronous API, per Anthropic's model page. The Batch API can reach 300K output tokens with the output-300k-2026-03-24 beta header.
Is Claude Opus 5 better than Fable 5?
Not across the board. Anthropic says Opus 5 comes close to Fable 5 at half the price. At max effort on CursorBench 3.2 it sits within 0.5% of Fable 5's best score. Anthropic still points developers to Fable for the hardest reasoning.