Anthropic released Claude Sonnet 5.5 just six days after unveiling Opus 5.5, positioning the former as a lower-cost alternative in its frontier model lineup. The company reports Sonnet 5.5 achieves 70.6% on Terminal-Bench 4.0, surpassing Opus 5.5's 66.4% at xhigh effort, while generating output more than 30% faster than its predecessor and consuming fewer tokens per task.

Pricing-wise, Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens—exactly half of Opus 5.5's $4 and $20 rates. However, lower per-token pricing does not automatically translate to proportionally lower bills. Artificial Analysis found that at maximum effort, Sonnet 5.5 actually costs $7.67 per task compared to Opus 5.5's $5.98, suggesting token efficiency varies by workload.

To evaluate whether Sonnet 5.5 merits adoption over Opus 5.5, three separate coding tasks were executed through the Anthropic API using identical prompts, adaptive thinking, and maximum effort settings. Each test ran five times per model to assess consistency, with results graded against hidden test suites the models had never encountered. Metrics tracked included token consumption, cost at list price, and execution time.

Test Design

  • Agentic bug fix: Models received a small Python order-pricing repository containing four intentional bugs and one flaky test, along with tools for file reading, writing, and test execution. A 12-test hidden suite evaluated the fixes.
  • Resolver spec: Models wrote a dependency resolver for a fictional package manager based on a two-page specification without executing code. A 120-test hidden suite graded the implementation.
  • Concurrency bugs: Models debugged an asyncio job queue containing three race conditions, informed by an incident report describing double charges and unexecuted jobs. Eight hidden tests validated each fix.

Results by Task

Agentic bug fix

Both models fixed all four bugs across all five runs and passed all 12 hidden tests without modifying test files. Sonnet 5.5 averaged 5 minutes 8 seconds, 29 tool calls, 42,608 output tokens, and $0.70 per run. Opus 5.5 averaged 3 minutes 21 seconds, 25 tool calls, 20,625 output tokens, and $0.75.

Sonnet 5.5 encountered difficulties with token limits during initial testing. The model's extended reasoning in a single step caused it to exceed the 32,000-token limit in four of five runs, halting execution before completion. Opus 5.5 remained well within the same constraint. After raising the limit to 128,000 tokens and rerunning the failed attempts, all passed. Counting the failed runs, Sonnet 5.5 averaged approximately $0.98 per run, exceeding Opus 5.5's $0.75.

Sonnet 5.5 thought so long in a single step that it hit that limit in four of five runs…

Opus 5.5 won this task. Both models achieved perfect fixes, but Opus 5.5 completed the work roughly 35% faster and, accounting for Sonnet 5.5's failed attempts, cost less overall.

Resolver spec

Sonnet 5.5 prevailed on this test. Both models passed all 120 hidden tests across all five runs, but Sonnet 5.5 achieved this faster and at lower cost. Sonnet 5.5 averaged 8 minutes 58 seconds, 81,097 output tokens, and $0.82 per run. Opus 5.5 averaged 9 minutes 40 seconds, 70,687 output tokens, and $1.42 per run. Despite consuming 15% more tokens, Sonnet 5.5 cost 42% less and finished slightly quicker.

Concurrency bugs

This test produced the most striking difference. Sonnet 5.5 fixed all three race conditions and passed all eight hidden tests on every run. Opus 5.5 succeeded on three runs but exhausted its 128,000-token limit on the other two without producing a solution.

Sonnet 5.5 averaged 12 minutes 13 seconds, 101,788 output tokens, and $1.02 per run. Opus 5.5 averaged 16 minutes 43 seconds, 111,428 output tokens, and $2.24 per run when including failures. Sonnet 5.5 decisively won this test, delivering perfect results every time while running faster and costing less than half as much.

Overall Performance

Across all 15 runs, Sonnet 5.5 achieved perfect results on every attempt, while Opus 5.5 succeeded on 13 of 15, with both failures occurring in the concurrency test. Total cost came to $12.69 for Sonnet 5.5 versus $22.07 for Opus 5.5—a 42% savings. Even accounting for the four runs that required adjustment, Sonnet 5.5 cost approximately $14.09, still roughly 36% less than Opus 5.5. Sonnet 5.5 also completed all 15 runs 11% faster overall. It won two of three tests: the resolver and concurrency tasks. Opus 5.5 won the agentic test, finishing about 35% faster while Sonnet 5.5 consumed twice as many tokens.

Interpretation

The findings diverge from Artificial Analysis's earlier assessment, which found Sonnet 5.5 more expensive per task than Opus 5.5 at maximum effort while scoring slightly lower on its Intelligence Index. In these tests, Sonnet 5.5 demonstrated superior accuracy and, even after accounting for rerun costs, delivered approximately 36% savings overall.

The per-token price advantage did not translate into a proportional reduction in total expenditure. Sonnet 5.5 frequently required additional tokens to complete tasks, resulting in a 36% overall cost reduction rather than the theoretical 50% savings from halved pricing.

For coding-intensive work, Sonnet 5.5 should serve as the default choice. It delivered flawless results on every run once its output token limit was increased, so setting this limit high is essential since it reasons longer per step than Opus 5.5. For agent loops and iterative workflows, Opus 5.5 remains preferable—it completed the agentic test roughly 35% faster and, when accounting for Sonnet 5.5's failed attempts, incurred lower costs.