What is Claude Opus 4.7?

Claude Opus 4.7 is a new hybrid reasoning AI model released by Anthropic, designed for improved honesty and reduced hallucination rates.

How does Claude Opus 4.7 compare to Claude Mythos?

Claude Opus 4.7 shows improvements in honesty and reduced hallucinations, but it is outperformed by the unreleased Claude Mythos, which Anthropic considers too dangerous for public release.

What are the safety improvements in Claude Opus 4.7?

Anthropic reports that Claude Opus 4.7 has reduced rates of hallucination, sycophancy, and important omissions, alongside moderate improvements in factuality.

Home / Technology / New Claude Model Slasher Hallucinations

New Claude Model Slasher Hallucinations

17 Apr

•

Summary

Claude Opus 4.7 shows reduced hallucination rates.
Model demonstrates improved honesty and reduced sycophancy.
It trails the unreleased, more advanced Claude Mythos.

Anthropic has released Claude Opus 4.7, a new hybrid reasoning AI model. This system card indicates improved performance in honesty and a reduced likelihood of hallucination and sycophancy compared to earlier Anthropic models and other frontier AI systems. Anthropic reports significant reductions in important omissions and moderate improvements in factuality. Opus 4.7 scores better on certain hallucination metrics than Gemini 3.1 Pro and Grok 4.20, according to Anthropic.

Claude Opus 4.7 is available to paid subscribers, differing from the unreleased Claude Mythos, which is shared with partners. Anthropic states Mythos is more advanced but too dangerous for public release due to cybersecurity capabilities. Opus 4.7 nevertheless enhances coding, visual intelligence, and document analysis over Opus 4.6. Hallucination rates are measured through various benchmarks like MASK, where Opus 4.7 showed a 91.7 percent honesty rate. Input hallucination rates remain high, around 90 percent, comparable to OpenAI's models.