AI Daily Brief — 16 April 2026

Anthropic ships Claude Opus 4.7 — a meaningful upgrade to its flagship: +10.9 points on SWE-bench Pro (64.3% vs 53.4% for Opus 4.6), higher-resolution vision (3.75 MP), file-system-level memory, and a new xhigh reasoning tier. Pricing stays at $5 / $25 per M tokens. Anthropic openly concedes Opus 4.7 still trails the unreleased Mythos.

Top stories

• Claude Opus 4.7 launches. The headline number: SWE-bench Pro 64.3% — up from 53.4% on Opus 4.6. Vision resolution jumps to 3.75 MP; file-system-level memory ships; a new xhigh reasoning tier joins low/medium/high. Pricing unchanged at $5 input / $25 output per million tokens. Available across Claude products, the API, and via Microsoft, Google and Amazon clouds. via Anthropic
• Opus 4.7 vs Mythos. Anthropic publicly concedes that Opus 4.7’s performance does not match Mythos — the Glasswing-only frontier model — held back because of its demonstrated cybersecurity capabilities. via Axios
• Built-in cybersec safeguards. Opus 4.7 ships with auto-detect-and-block on prohibited or high-risk cyber requests — a direct response to the Mythos-era safety findings. via CNBC
• Bedrock support day-one. Anthropic and AWS make Opus 4.7 available in Amazon Bedrock at launch. via AWS

By the numbers

• SWE-bench Pro: 64.3% on Opus 4.7 vs 53.4% on Opus 4.6 (+10.9 pp)
• Vision: 3.75 MP
• Pricing: $5 / $25 per M tokens (unchanged)
• Reasoning tiers: low · medium · high · xhigh (new)

Compiled by AI Feed’s editor from publicly reported announcements on 16 April 2026.