Claude Opus 4 and Claude Sonnet 4

Training data disclosure: Vague

Level determined for 0% of benchmark cells.

• TestTest data seen0
• SeenSeen, split unclear0
• TrainTraining data seen0
• Rel.Related data seen0
• NoNot seen0
• ?Unknown22

Other declared training data

• publicly available information on the Internet as of March 2025 (broad web data: may contain benchmark test sets) · split: unknown
  Evidence

  “Claude Opus 4 and Claude Sonnet 4 were trained on a proprietary mix of publicly available information on the Internet as of March 2025, as well as non-public data from third parties, data provided by data-labeling services and paid contractors, data from Claude users who have opted in to have their data used for training, and data we generated internally at Anthropic.”

• non-public data from third parties (not yet matched to a dataset) · split: unknown
  Evidence

  “Claude Opus 4 and Claude Sonnet 4 were trained on a proprietary mix of publicly available information on the Internet as of March 2025, as well as non-public data from third parties, data provided by data-labeling services and paid contractors, data from Claude users who have opted in to have their data used for training, and data we generated internally at Anthropic.”

• data provided by data-labeling services and paid contractors (not yet matched to a dataset) · split: unknown
  Evidence

  “Claude Opus 4 and Claude Sonnet 4 were trained on a proprietary mix of publicly available information on the Internet as of March 2025, as well as non-public data from third parties, data provided by data-labeling services and paid contractors, data from Claude users who have opted in to have their data used for training, and data we generated internally at Anthropic.”

• data from Claude users who have opted in to have their data used for training (not yet matched to a dataset) · split: unknown
  Evidence

  “Claude Opus 4 and Claude Sonnet 4 were trained on a proprietary mix of publicly available information on the Internet as of March 2025, as well as non-public data from third parties, data provided by data-labeling services and paid contractors, data from Claude users who have opted in to have their data used for training, and data we generated internally at Anthropic.”

• data we generated internally at Anthropic (not yet matched to a dataset) · split: unknown
  Evidence

  “Claude Opus 4 and Claude Sonnet 4 were trained on a proprietary mix of publicly available information on the Internet as of March 2025, as well as non-public data from third parties, data provided by data-labeling services and paid contractors, data from Claude users who have opted in to have their data used for training, and data we generated internally at Anthropic.”

Benchmark cells

Classic LLM benchmarks

Multimodal LLM benchmarks

Recent LLM benchmarks

Notes

Reliable knowledge cutoff January 2025 (Transparency Hub). The system card says Anthropic is adding canary strings to its own published transcripts to help exclude them from future pretraining corpora. No benchmark decontamination statement found. Other reported contamination, not shown as cells: Alignment Faking in Large Language Models (publicly released transcripts): ~150,000 publicly released transcripts from the 'Alignment Faking in Large Language Models' paper were in Claude Opus 4's pretraining data; related evaluations could not be used reliably Accepts image input: “Claude Opus 4 and Claude Sonnet 4 can understand both text (including voice dictation) and image inputs, engaging in conversation, analysis, coding, and creative tasks.”

Sources

• System Card: Claude Opus 4 & Claude Sonnet 4 · model card · retrieved 2026-10-15
• Anthropic Transparency Hub: Model Report · website · retrieved 2026-10-15
• Claude docs: Model deprecations (model IDs) · website · retrieved 2026-10-15
• Introducing Claude 4 · website · retrieved 2026-10-15