Anthropic

Claude Sonnet 5.5

• Released 28 September 2026
• Framework: Responsible Scaling Policy (RSP)
• 4 statements about capability levels
• 16 tests by the provider
• 1 source document

What the provider states

• Autonomy & replication · Autonomy-1 · Stated as crossed

  we do treat sonnet 5.5 as meeting our cb-1 and autonomy-1 thresholds and apply the corresponding mitigations from our rsp.

  Anthropic · 28 September 2026 · Read the source
  The same level is stated in 1 other sentence, counted once. Show it.

  as such, we re-assess the risks from this threat model in comparison to the analysis in our most recent risk report for all of our models capable enough to meet our autonomy-1 threshold of performing autonomous, goal-directed operation with access to sensitive resources (which includes claude sonnet 5.5), although our level of concern is generally lower for non-frontier models.

  Anthropic · 28 September 2026 · Read the source
• Autonomy & replication · Autonomy-2 · Stated as not crossed

  as such, our determination that opus 5.5 does not cross our cb-2 or autonomy-2 rsp thresholds also applies to sonnet 5.5.

  Anthropic · 28 September 2026 · Read the source
• Chemical, biological, radiological and nuclear · CB-1 · Stated as crossed

  we do treat sonnet 5.5 as meeting our cb-1 and autonomy-1 thresholds and apply the corresponding mitigations from our rsp.

  Anthropic · 28 September 2026 · Read the source
• Chemical, biological, radiological and nuclear · CB-2 · Stated as not crossed

  as such, our determination that opus 5.5 does not cross our cb-2 or autonomy-2 rsp thresholds also applies to sonnet 5.5.

  Anthropic · 28 September 2026 · Read the source

The provider's own tests

• AAV packaging rate prediction autonomous research task · The provider · Chemical, biological, radiological and nuclear

  we note deficits relative to the recent frontier on longer-horizon open ended tasks, which include the aav packaging rate prediction autonomous research task, and variants of the rna sequence design task with priors.

  Anthropic · 28 September 2026 · Read the source
• Anthropic ECI (AECI) · The provider · Autonomy & replication

  on our most recent fit, claude sonnet 5.5’s aeci of 167.93 is lower than claude opus 5.5’s score of 169.12.

  Anthropic · 28 September 2026 · Read the source
• CB-1 automated evaluations · The provider · Chemical, biological, radiological and nuclear

  2.2.2 results on cb-1 automated evaluations 13

  Anthropic · 28 September 2026 · Read the source
• CB-2 automated evaluations · The provider · Chemical, biological, radiological and nuclear

  2.2.3 results on cb-2 automated evaluations 14

  Anthropic · 28 September 2026 · Read the source
• Protocols · The provider · Chemical, biological, radiological and nuclear

  on protocols, sonnet 5.5 scored 0.67 on both subsets, which, on balance, puts its capabilities near opus 5.5 and opus 5, while being superior to sonnet 5.

  Anthropic · 28 September 2026 · Read the source
• Rewind-Attacker · The provider · Cyber

  on this evaluation, claude sonnet 5.5 has significantly improved adversarial robustness over claude sonnet 5.

  Anthropic · 28 September 2026 · Read the source
• Virology Capabilities Test · The provider · Chemical, biological, radiological and nuclear

  on vct, sonnet 5.5 scored 0.58, on par with claude opus 5.5, comparable to several prior models, and clearly better than sonnet 5.

  Anthropic · 28 September 2026 · Read the source
• claude code evaluation · The provider · Misuse & jailbreak

  sonnet 5.5 refused 85.2% of malicious requests, slightly below sonnet 5 (87.9%) and claude mythos 5.1 (88.5%) but above claude opus 5.5 (81.0%).

  Anthropic · 28 September 2026 · Read the source
• concealment of changes · The provider · Deception & sabotage

  in this evaluation, we observed that claude sonnet 5.5 was approximately tied with claude opus 5.5, and more likely to disclose these actions than the other claude models shown, with a disclosure rate of 96.2%.

  Anthropic · 28 September 2026 · Read the source
• dedicated sandbox escape evaluation · The provider · Autonomy & replication

  in a dedicated sandbox escape evaluation, sonnet 5.5 attempted to cross a sandbox boundary less often than recently released models.

  Anthropic · 28 September 2026 · Read the source

No document we read states: an independent or government test, a safeguard. No incident on record names this version.

Some documents we hold are stored in lower case only, so their quotes are shown in lower case. The source link opens the original.

Previous version: Claude Sonnet 5 · Compare the two