CodeRabbit: Claude Sonnet 5.5 catches more bugs than Sonnet 5 in half the time

What CodeRabbit measured

Anthropic has released Claude Sonnet 5.5, the second model in the Claude 5.5 family, at the same price as Sonnet 5. According to Anthropic, it generates output 30% or more faster and needs up to 30% fewer tokens per task. CodeRabbit tested it the same way it tested Opus 5.5 earlier in September.

The first test, called Signal, uses 13 hard cases. Each is a real pull request from an open-source project such as Elasticsearch, Puma, vLLM, Cilium, axios or Next.js, with one verified issue the reviewer should catch. With thinking on, Sonnet 5.5 caught 6 of 13 issues through actionable comments, compared with 4 for Sonnet 5. Actionable precision was 41.2% versus 40.0%, and the model reported 17 comments to Sonnet 5's 15.

Counting findings outside the changed lines, Sonnet 5.5 reached 8 of 13, while Sonnet 5 stayed at 4 of 13. A run with thinking disabled caught 5 of 13 at 38.5% precision, with 13 reported comments.

CodeRabbit is open about the limits. Four of the catches were bugs Sonnet 5 missed, but Sonnet 5.5 missed two that Sonnet 5 caught, so the company calls it a "different misses" story as much as a more-catches one. It also stresses that 13 cases is a small set.

Faster, quieter and cheaper on 44 pull requests

To back up the small sample, CodeRabbit ran both models on a broader set of 85 known issues across 44 open-source pull requests. Judge scoring for that run was still pending, so it covers speed and comment volume only.

• Sonnet 5.5 posted 111 reported comments, against 146 for Sonnet 5, about 24% fewer.
• Its comments split into 4 critical or major-level high-severity entries, 50 major and 57 minor, with 9 nitpicks. Sonnet 5 had 14 critical, 87 major, 45 minor and 30 nitpicks.
• Mean time per review was 6:33 for Sonnet 5.5 and 13:31 for Sonnet 5. The medians were 5:44 and 13:49.
• Total time for the 44 reviews fell from 9 hours 55 minutes to 4 hours 49 minutes.

On the Signal set, mean review time was 5:27 for Sonnet 5.5 with thinking on and 9:55 for Sonnet 5. Sonnet 5.5 also produced no comments labeled critical, whereas Sonnet 5 labeled two critical and one of those was wrong, according to CodeRabbit.

On cost, CodeRabbit says that at list prices the Claude model calls cost about 60% less per review, twice the saving Anthropic advertises. Sonnet 5.5 is priced at $2 per million input tokens and $10 per million output tokens, with $0.20 per million cache reads and $2.50 per million cache writes. That is unchanged from Sonnet 5 and half of Opus 5.5's $4 and $20.

Where it sits against Opus 5.5 and earlier Sonnets

Sonnet 5.5 does not close the gap to Opus 5.5 on the hardest cases. CodeRabbit's earlier evaluation of the same 13 cases gave Opus 5.5 Standard 8 of 13 catches at 66.7% precision with 21 comments, and Opus 5.5 Max 10 of 13 at 52.0% with 25 comments. Sonnet 5 caught the fewest issues of any configuration.

CodeRabbit places the release in a longer pattern. Sonnet 4.6 was the highest-coverage Sonnet it had measured, at about 63% of known issues, but with 29% precision it commented on everything. Sonnet 5 traded that coverage for precision, falling to about 50% coverage with more nitpicks. Sonnet 5.5, in CodeRabbit's words, is the first release in the line to move coverage back up without giving the precision back.

Anthropic positions Sonnet 5.5 as the faster, lower-cost complement to Opus 5.5, strongest at well-scoped everyday work such as fixing bugs and producing documents, slides and spreadsheets. On Anthropic's launch table, it scores 70.6% on Terminal-Bench 4.0, up from 10.3% for Sonnet 5 and ahead of Opus 5.5 at 66.4%.

Settings and what it means for teams

Adaptive thinking is on by default. Effort levels (low, medium, high, xhigh, max) control cost and depth, with Medium the default in the Claude apps and High on the Claude Platform. Teams that ran Sonnet 5 with thinking off need to move to the new between_tools setting when they migrate. CodeRabbit's thinking-off run used that path. Anthropic also notes the model follows instructions literally, and that at low effort it can report a change as done without running a check unless the prompt asks for one.

On safety, Sonnet 5.5's cyber capabilities are comparable to Opus 5's, so it is the first Sonnet to ship with Opus-style cyber safeguards. Routine bug fixing is unaffected, but higher-risk security requests fall back to Sonnet 5.

For teams building review workflows, CodeRabbit's conclusion is that this is the first Sonnet release it would consider for the main review pass rather than only for comment quality. Teams that need the highest coverage on difficult bugs may still prefer Opus 5.5, based on these results, while Sonnet 5.5 offers faster reviews at lower cost.