Descriptions:
Nick Saraev offers a benchmark-focused breakdown of Claude Opus 4.7 alongside his broader read on where the model sits in Anthropic’s lineup and what it signals about the near-term competitive landscape. Working from Anthropic’s official benchmark scorecard, Saraev walks through the numbers with specific figures: SWE Pro climbs from 53.4% to 64.3%, visual reasoning jumps from 69.1% to 82.1%, and Humanity’s Last Exam improves from roughly 40% to 46.9% — with Mythos Preview still significantly ahead at 56.8% on that last metric.
Saraev’s central interpretation is that Opus 4.7 lands approximately halfway between Opus 4.6 and Mythos Preview across most benchmarks, which he reads as intentional: a capable model released while Anthropic retains the most powerful capabilities in the unreleased Mythos. He flags two categories where 4.7 underperforms 4.6 — Agentic Search (Browse Comp) and cybersecurity vulnerability reproduction — and argues these regressions are likely deliberate safety-motivated throttling rather than genuine capability limitations, given their alignment with the security concerns that kept Mythos from public release.
The video closes with a prediction that most current benchmarks will be saturated within one model generation, and that OpenAI’s rumored “Spud” model will drop within days of Opus 4.7’s release. For developers and researchers tracking the frontier model competitive landscape, Saraev provides a concise, numerically grounded orientation to where Opus 4.7 actually sits.
📺 Source: Nick Saraev · Published April 16, 2026
🏷️ Format: Review
