Claude Opus 5.5 Launched: Fable Level AI at 40% Lower Cost
Anthropic launches Claude Opus 5.5, matching Fable 5.1 on most work while costing 40 percent less and running faster.
American artificial intelligence company Anthropic has launched Claude Opus 5.5, its newest flagship AI model. The model was released on Tuesday, September 22, 2026, and is now available to users around the world. Anthropic says it performs at the level of its more expensive Claude Fable 5.1 model on most work, while costing about 40 percent less to run than the older Opus 5.
Opus 5.5 is the first model in the new Claude 5.5 family. It replaces Opus 5, which had been on the market for only about two months. The launch came on the same day that rival company OpenAI unveiled its own new models, which shows how fast the AI race is moving.
Beating Fable 5.1 on Most Tests
According to the benchmark results shared by Anthropic, Opus 5.5 scored higher than Fable 5.1 in coding, office work, computer use, reading charts and difficult reasoning questions. On Terminal-Bench 4.0, a test of complex computer tasks, Opus 5.5 scored 66.4 percent. Fable 5.1 scored 55.8 percent and Opus 5 scored 52.3 percent.
On a test of real professional work across 44 jobs, called GDPval-AA, Opus 5.5 also came first among Claude models. However, Anthropic itself admits that test scores do not tell the full story. The company says that in its own daily use, the gap between Opus 5.5 and Fable 5.1 is smaller than the numbers suggest.
OpenAI's new models still lead in some areas. On a test of business workflows, OpenAI's GPT-6 Astra scored 41.4 percent against 40.0 percent for Opus 5.5. It also scored higher on a test of scientific research tasks, with 64.6 percent against 58.7 percent.
Big Jobs Finished in Hours
Anthropic says early testers saw large jumps in performance on their hardest work. One tester used Opus 5.5 to move a codebase of 680,000 lines to a new system in less than a day. The company says an engineering team would have needed weeks for the same job.
In another test, Anthropic asked the model to make every page of a web app load faster. Opus 5.5 succeeded in 39 out of 40 tries. Opus 5 made smaller improvements in the same test and also changed how the app behaved, which is something developers do not want.
The company also asked Opus 5.5 and Fable 5.1 to rewrite HAProxy, a popular piece of server software, from the C language into Rust. Opus 5.5 finished in 9.5 hours, while Fable 5.1 took 12 hours. Anthropic says Opus 5.5 did the job at 51 percent lower cost. In a separate game building test, Opus 5.5 got the highest score of all Claude models for graphics and polish.
Lower Price and Faster Answers
The biggest news for businesses is the price. Opus 5.5 needs less computing power than Opus 5, so Anthropic has cut its prices by 20 percent. Cache reads, which make up most of the cost in coding and agent work, are now 60 percent cheaper.
Price per 1 million tokens | Claude Opus 5.5 | Claude Opus 5 |
|---|---|---|
Input tokens | $4 | $5 |
Output tokens | $20 | $25 |
Cache reads | $0.20 | $0.50 |
Cache writes | $5 | $6.25 |
Tokens are small pieces of words that AI companies use to measure usage and charge customers. Because the new model also uses fewer tokens to finish each task, Anthropic says the total cost of a typical workload falls by about 40 percent. For comparison, Fable 5.1 costs $10 per million input tokens and $50 per million output tokens, which is more than double.
Opus 5.5 also writes its answers more than 30 percent faster than Opus 5. Anthropic has raised the five hour usage limits on its Pro, Max, Team and seat based Enterprise plans. Subscribers have also received a one time usage limit reset, which reports say can be used any time until October 22.
Safety Scores and New Safeguards
Anthropic says Opus 5.5 got the best results of any of its models so far on its automated behavioral audit. This test checks how the AI behaves in nearly 2,000 simulated situations. Outside groups, including METR and Frontier Design, also tested the model before release.
The company says Opus 5.5 is less likely to take actions that are hard to undo, or to go beyond the limits set for it. In one new test, it tried to cross those limits about 85 percent less often than Opus 5. It is also better at resisting prompt injection, a trick where hidden instructions are used to fool an AI into doing something harmful.
Anthropic has been open about the limits of its testing. It says there are signs that the model often suspects when it is being tested, which makes it harder to know how it will act in the real world. Full details are in the company's system card.
Because Opus 5.5 is very strong in biology and cybersecurity, Anthropic is using safeguards similar to those on Fable 5.1. Most risky cybersecurity requests will be sent to an older Claude model instead. Approved research groups can apply to the Life Sciences Verification Program for biology work, and verified cybersecurity experts will get access through the Cyber Verification Program in the coming weeks.
What It Means for Pakistan
For Pakistan, cheaper AI tools matter a lot. Pakistani software houses and freelancers pay for these services in US dollars, so every price cut makes a real difference to their costs. Pakistan based freelancers earned about $400 million in the first nine months of the 2025 financial year, and many of them now use AI tools for coding, writing and design work.
A 40 percent drop in running costs could help small Pakistani startups build AI products that were too expensive before. It could also help developers who build chatbots and apps for foreign clients offer lower prices and finish work faster.
The launch also connects with the government's own plans. In July 2025, the federal cabinet led by Prime Minister Shehbaz Sharif approved the National AI Policy 2025. The policy aims to train one million AI professionals by 2030 and offer 3,000 AI scholarships every year. Tools that do more work at a lower price may help students and young workers learn these skills.
There is another side too. If AI can finish in hours what took a team weeks, some basic coding and data jobs that Pakistani freelancers depend on may face more pressure. Workers who learn to use these tools well are likely to be in a stronger position than those who do not.
Sonnet 5.5 and Haiku 5.5 Coming Soon
Opus 5.5 is available now on the Claude apps and website, and through Amazon Web Services, Google Cloud and Microsoft Azure. Developers can find full details on Anthropic's official announcement page.
Anthropic says two smaller models, Claude Sonnet 5.5 and Claude Haiku 5.5, will launch in the coming weeks. The company expects them to bring many of the same gains in performance, speed and safety. Exact dates and prices for these models have not been announced yet.