Anthropic

Claude Sonnet 5

• Released 30 June 2026
• Framework: Responsible Scaling Policy (RSP)
• 2 statements about capability levels
• 10 tests by the provider
• 6 source documents

What the provider states

• Chemical, biological, radiological and nuclear · CB-2 · Stated as not crossed

  sonnet 5 does not cross the cb-2 threshold.

  Anthropic · 30 June 2026 · Read the source
• Chemical, biological, radiological and nuclear · Notable capability · Not determined

  we consider performance on the virology tasks notable if: ● the model scores above 0.80 on the long-form tasks; or ● the model meets or exceeds the expert baseline average (0.221) on the multimodal task.

  Anthropic · 30 June 2026 · Read the source

The provider's own tests

• BioMysteryBench · The provider · Chemical, biological, radiological and nuclear

  8.14.1 biomysterybench biomysterybench assesses a model’s ability to solve difficult, analytical challenges that require interleaving computational analysis with biological reasoning.

  Anthropic · 30 June 2026 · Read the source
• CB-1 automated evaluations · The provider · Chemical, biological, radiological and nuclear

  2.2.4 biological risk results: cb-1 automated evaluations 14

  Anthropic · 30 June 2026 · Read the source
• CB-2 automated evaluations · The provider · Chemical, biological, radiological and nuclear

  2.2.5 biological risk results: cb-2 automated evaluations 16

  Anthropic · 30 June 2026 · Read the source
• ProteinGym hard · The provider · Chemical, biological, radiological and nuclear

  8.14.4 proteingym hard the proteingym benchmark assesses a model’s ability to predict the effects of protein mutations.

  Anthropic · 30 June 2026 · Read the source
• automated offline monitoring pipeline · The provider · Deception & sabotage

  our automated offline monitoring did not surface any instance of sandbagging, overtly malicious actions, long-horizon strategic deception, or long-horizon strategic oversight evasion.

  Anthropic · 30 June 2026 · Read the source
• influence operations testing · The provider · Misuse & jailbreak

  in influence operations testing, for example, the model sometimes grounded its refusals in practical concerns (i.e., why the plan wouldn’t succeed) rather than in principled ones (i.e., why the request was potentially harmful).

  Anthropic · 30 June 2026 · Read the source
• live bug bounty · The provider · Injection & agentic

  sonnet 5 tied with claude opus 4.8 for the strongest result in the bug bounty, with only 0.19% of unique attacks succeeding against each, a significant improvement over claude sonnet 4.6 (1.41%) and other frontier models like gpt-5.5 (3.08%) or gemini 3.5 flash (6.66%).

  Anthropic · 30 June 2026 · Read the source
• organic chemistry · The provider · Chemical, biological, radiological and nuclear

  8.14.5 organic chemistry we evaluated models’ fundamental skills spanning tasks like predicting molecular structures from spectroscopy data, designing multi-step synthetic routes, predicting reaction products, and converting between iupac names, smiles notation, and chemical structure images.

  Anthropic · 30 June 2026 · Read the source
• protocol troubleshooting · The provider · Chemical, biological, radiological and nuclear

  8.14.6 protocol troubleshooting this assessment looks at models’ ability to detect and fix errors in molecular biology protocols, including by using web search tools to find additional details about protocols online.

  Anthropic · 30 June 2026 · Read the source
• structural biology, open-ended · The provider · Chemical, biological, radiological and nuclear

  8.14.3 structural biology, open-ended we evaluated the model’s ability to understand the relationship between biomolecular structure and function.

  Anthropic · 30 June 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 4.6 · Compare the two