Opus 5.5 ported HAProxy to Rust in 9.5 hours, Anthropic says
Anthropic's own head to head has the new model finishing a real port faster and cheaper, which is exactly why it needs an outside run.
Anthropic launched Claude Opus 5.5, the first model in what it calls the Claude 5.5 family, and says it performs at the level of Claude Fable 5.1 for most tasks while costing 40% less to run than Opus 5.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.
The number founders will actually care about came from Boris Cherny, who posted the head to head. Anthropic had Opus 5.5 and Fable 5.1 each port HAProxy, a load balancer written in C, over to Rust. Both models passed nearly all of HAProxy's tests. Opus 5.5 finished in 9.5 hours against Fable 5.1's 12, and did it for 51% less cost. Cherny says the model has been his daily driver for the last few weeks.
Opus 5.5 finished in 9.5 hours compared to Fable 5.1's 12 hours, and for 51% less cost
Why the task matters more than the percentage
Porting a mature C codebase to Rust is not a toy benchmark. It is long, boring, unglamorous work with a hard pass or fail at the end, which is the shape of the work a solo founder would most like to hand off. A model that can run for hours on that kind of job and come back with something that passes the existing test suite is a different proposition to one that writes a nice function when you ask nicely.
The cost line is the part worth sitting with. Two models, same job, both passing, one of them 51% cheaper and two and a half hours faster. If you are paying for tokens on long agent runs, that gap compounds every single time you kick off a task, and it shows up on the card at the end of the month rather than in a demo.
The business read
This is Anthropic benchmarking Anthropic on a task Anthropic picked, reported by an Anthropic employee. Nothing in the sources says otherwise, and nobody outside the company has run the same port and published numbers yet. Treat the 9.5 hours and the 51% as a vendor claim, the same way you would treat any founder posting their own screenshot.
The honest move if the cost claim matters to your margins is to take a job you already run repeatedly, run it on both, and compare your own bill. That is a couple of hours of work and it replaces a marketing number with a number you own.
What the launch does tell you cleanly is direction. Anthropic is pitching this release on price per unit of work rather than on raw capability, which is a reasonable signal about where it thinks buyers are now.