Claude Sonnet 4.6

User comments (primarily from HackerNews, 104 total) paint a nuanced picture. On the POSITIVE side: a mocha team ran zero-shot app-building experiments and found Sonnet 4.6 'ran up to 3-4x longer than Sonnet 4.5 without intervention, producing functional apps on par in terms of quality to the Opus series,' calling it 'a fundamentally different model' closer to Opus in autonomy. Another user had Claude implement 'a working web-based email client from scratch in Rust which can interact with a JMAP based mail server' in about 20 minutes, with only minor bugs requiring follow-up prompting. Multiple users discuss how the model's coding ability is good enough to disrupt SaaS — if you can generate task-specific software, you may not need to buy it. On the CONCERN side: one user measured a '15-45% increase in output token amount compared to 4.5,' particularly on complex inference tasks — meaning higher API costs. Another user demonstrated that the model 'has no semantic understanding' by testing the classic bat-and-ball problem, finding it 'pretended to do some algebra' rather than genuinely reasoning. Security researchers flagged that agentic LLM use reintroduces 'in-band signaling' vulnerabilities — prompt injection via external data sources is a live attack vector. Broader discussion centered on labor displacement: users note this tech 'can cause one engineer to do the work of 3,' with companies likely to freeze hiring rather than increase output. Several users who work with local models (4B-30B params) noted they 'actively avoid cloud-based LLMs,' highlighting a divide between those who can afford frontier models and those who cannot.

No aggregate ratings were found for this product during the last harvest.