Anthropic Releases Claude Opus 4.7: Benchmarks, Vision Upgrades, Memory, Pricing & New Claude Code Features
Anthropic has released Opus 4.7, and the video covers the announcement, benchmark results, and related product updates. Opus 4.7 shows significant improvements in software engineering and is described as more tasteful and creative for professional tasks like front-end interfaces, slides, and documentation, while still trailing the newly announced Claude Mythos preview model. Benchmarks show a notable jump from Opus 4.6 to 4.7, including over a 10-point gain on SWE Bench Pro, plus better vision-related performance for agent computer use and visual reasoning. Opus 4.7 is available via the API, Claude Code, web, and desktop apps, with pricing unchanged at $5/M input tokens and $25/M output tokens. The release also highlights deeper, less sycophantic responses, improved file-system-based memory across multi-session work, a new API “extra high effort” parameter, beta task budgets, a Claude Code “ultra review” command, and expanded auto mode to Max users, alongside a note that Opus 4.7 may use substantially more tokens at Max effort.
00:00 Opus 4.7 Overview
00:10 Coding And Creativity Gains
00:44 Benchmark Highlights
01:14 Vision And Web Use
01:30 Availability And Pricing
01:46 Less Sycophancy Feedback
02:22 File Based Memory
03:09 New API Controls
04:00 Claude Code Updates
04:22 Effort Level Token Costs
04:50 Wrap Up And Outro
source
