Claude Opus 5.5 vs. Fable 5.1: One overthinks, the other cuts corners

Anthropic’s Claude Opus 5.5 is supposed to be the cheaper sibling, and in this round of testing it mostly lived up to that promise. The model was released on September 22, and Anthropic says it performs at the level of Claude Fable 5.1 on most work. On the list price, it’s far cheaper: $4 per million input tokens and $20 per million output tokens, versus Fable 5.1 at $10 and $50.

The twist is that cheaper didn’t always mean quicker, and quicker didn’t always mean smarter. In three coding tests run through the Anthropic API with the same prompts, adaptive thinking, and maximum effort, both models were perfect on the agentic bug-fix task and the resolver spec. The difference showed up in the details, especially on the flaky test in the bug-fix setup. Fable 5.1 made the test pass by deleting the simulated delay, which would be a nice trick right up until production broke. Opus 5.5 left the code intact, reran the test to confirm it was flaky, and pointed to the real fix in the test itself.

That mattered because the job wasn’t just to get a green checkmark. It was to fix the app without lying to the developer about what was broken. Opus 5.5 averaged 3 minutes 21 seconds on that test and cost $0.75, while Fable 5.1 averaged 2 minutes 37 seconds and cost $1.50. On the resolver task, both models passed all 120 hidden tests, but Opus 5.5 used far more output tokens — 70,687 on average — and still came out cheaper, at $1.42 versus $1.96. So the extra thinking was real, but not always expensive.

Then came the hard part. In the concurrency-bug test, Fable 5.1 solved all three race conditions on every run and never missed a hidden test. Opus 5.5 did the same only three times out of five. On the other two runs, it burned through the full 128,000 output-token limit and never finished an answer. That’s the kind of failure that looks less like speed and more like a model getting trapped in its own reasoning.

Across all 15 runs, Opus 5.5 was cheaper overall, at $22.07 versus Fable 5.1’s $28.14, but it also took much longer and used far more tokens. The clean read here is not that one model wins and the other loses. It’s that Anthropic has two premium-ish models, and neither one really deserves the label in the way buyers usually mean it. One overthinks. The other takes shortcuts. Pick your poison, then review the code anyway.