Claude Fable 5.2 vs. Opus 5.5: Same scores at almost 4x the cost

Anthropic released Claude Fable 5.2 on November 9, claiming it outperforms Claude Opus 5.5 in almost every capability the company measured. The company suggested using Opus for most coding and knowledge tasks, positioning it as its "strongest daily driver." However, Fable 5.2 is significantly cheaper, which raises questions about whether the cheaper option is sufficient.

In coding benchmarks, Fable 5.2 scored 49.1% at Xhigh effort on FrontierCode 2.0, compared to 45.5% for Opus 5.5 at Max. Fable 5.2 also achieved 45.3% at High effort, roughly half the cost per task of Opus. On FrontierSWE v2, Fable scored 69.9%, while Opus scored 62.3%. While a price difference of twice as much should indeed make a meaningful impact, the reporter decided to put the models to the test by running both through Anthropic's API to examine their performance, token usage, cost, and execution time for coding and agent jobs.

The reporter conducted three different coding and agent jobs, measuring each model's performance, token usage, cost, and time taken. The tests were conducted on Anthropic's API without any adjustments or turn caps. The reporter used a script to grade each model's answers for the tests. For billing, Anthropic charges separately for uncached input, cache reads, cache writes, and output.

The reporter ran five tests: a cron scheduler to assess the model's ability to turn a detailed specification into correct code and adapt to changes without breaking it; an invoicing repository fix to test the model's ability to identify and fix bugs in a small invoicing library; and an incident investigation to see which model was better at debugging a payments outage from logs, metrics, and deployment records.

The reporter found that Fable 5.2 and Opus 5.5 tied with perfect scores on the first version of every test, but the reporter made each test tougher, rerunning them twice.

In the first test (cron scheduler), Fable 5.2 wrote significantly more reasoning than Opus 5.5 on every version, but the extra reasoning didn't improve Fable's score. Opus 5.5 finished each version in under a minute and cost about a third of what Fable 5.2 did. In the invoicing bug hunt test, both models scored perfectly on all versions, but Opus 5.5 used fewer reasoning tokens and completed the tasks faster, costing much less.

Lastly, in the incident investigation test, both models scored perfectly, but Opus 5.5 was quicker and used fewer reasoning tokens, resulting in a lower cost.

In conclusion, while the cheaper Claude Fable 5.2 performs comparably to the more expensive Claude Opus 5.5 in many tasks, the cost difference is significant and may not justify the choice for all users.