SYSTEM Cited by 2 sources
Claude Sonnet 4¶
Claude Sonnet 4 is Anthropic's frontier-tier Sonnet-class model, positioned as the middle-capacity rung of the Claude family (between Haiku and Opus). Production-available on AWS Bedrock and Anthropic's direct API.
Pattern of appearance in the wiki¶
Sonnet-4 shows up on the wiki as the default frontier-tier choice for production LLM workloads that need better accuracy than small open-source models but don't need Opus-class capability, typically hosted on AWS Bedrock when compliance review is a pre-condition for sending regulated text (PII, business metrics) to a third-party LLM.
Seen in¶
- sources/2025-09-24-zalando-dead-ends-or-data-goldmines-ai-powered-postmortem-analysis — Zalando's current-generation postmortem analysis pipeline runs on Claude Sonnet 4 on AWS Bedrock. Per-document processing time ~30 s. Hallucination rate "negligible" at this tier after prompt hardening. Residual surface attribution error rate disclosed at "approximately 10%, even with advanced models such as Claude Sonnet 4" — canonical wiki datum that attribution error is a structural reasoning limitation rather than a scale problem.
- sources/2026-05-18-cloudflare-project-glasswing-what-mythos-showed-us — post-Sonnet-4 Anthropic capability tier reference. Cloudflare's writeup explicitly benchmarks Mythos Preview (Anthropic's preview cyber frontier model delivered through Project Glasswing) against GA-tier general-purpose frontier models, naming "Opus 4.7 or GPT-5.5" as the comparison points the Glasswing-provided model was intentionally provided without the safeguards of. First wiki disclosure of an Anthropic Opus-4.7-tier model and of a cyber-specialised frontier model from Anthropic.
Related¶
- systems/amazon-bedrock — the hosting platform.
- systems/zalando-postmortem-analysis-pipeline — canonical production consumer.
- systems/mythos-preview — Anthropic's cyber-specialised frontier-tier preview model delivered through Project Glasswing.
- systems/anthropic-project-glasswing — the controlled- research program for Mythos Preview.
- concepts/cyber-frontier-model — the model class Mythos Preview belongs to (distinct from Sonnet's general-purpose positioning).
- concepts/llm-hallucination — the failure-mode category Sonnet-4 largely eliminates at this tier.
- concepts/surface-attribution-error — the failure mode Sonnet-4 does not fully eliminate (~10% residual).