The New Claude Opus 4.7 Can Actually Do This Now

Summary
Saj from Skill Leap AI reviews and tests Anthropic’s newly released Claude Opus 4.7 model. Through hands-on demonstrations in the Claude web interface, he benchmarks its coding, reasoning, vision, and long-context capabilities by generating interactive Three.js graphics, dashboards, animations, and web applications.

What is shown

• UI & Architecture Overview [00:00–03:28]: Demonstrates model selector showing Opus 4.7 with "Adaptive thinking," reviews benchmark charts comparing Opus 4.7 against Opus 4.6, GPT-5.4, Gemini 3.1 Pro, and the unreleased Mythos Preview, and details API pricing and effort controls (xhigh).
• Interactive 3D Isometric City Builder [04:06–05:40]: Prompts Opus 4.7 to generate an interactive Three.js city simulation with traffic, roads, and buildings, then tests the same prompt in Sonnet 4.6, which threw a generation error.
• Interactive AI Tools Comparison Website [05:40–07:30]: Prompts Opus 4.7 to build "Compare.ai," an interactive tool comparison web app featuring tool cards, filter tags, side-by-side comparison tables, working external URLs, and a dark/light mode toggle.
• 75 Years of AI Timeline [07:31–08:10]: Tests Opus 4.7 with adaptive thinking turned off on a simple prompt to build an interactive decadal timeline of AI history.
• Cinematic Global Metrics Visualization [08:11–09:23]: Generates a 3D animated globe visualizing population, CO2, life expectancy, and GDP per capita from 1820 to 2026; fixes a video playback bug using a single follow-up prompt.
• Photorealistic 3D Earth [09:24–10:06]: Creates an interactive Earth visualization using NASA textures, showing day/night illumination cycles and clickable city information pins (Chicago, New York, Los Angeles, Berlin).
• Interactive "Powers of Ten" Experience [10:07–10:40]: Renders a scroll-driven 3D Three.js animation zooming from a human on a picnic blanket down to quantum foam and out to cosmic scales.
• Image-to-Interactive Infographic [10:41–11:55]: Feeds a complex, cluttered "History of the World" image into Opus 4.7 to re-render it as a clean, interactive timeline dashboard spanning ~1,500 lines of code.
• Copyright Guardrail Test [11:56–12:11]: Tests prompting Opus 4.7 to recreate the copyrighted Pokémon Red opening sequence; the model refuses on IP grounds and suggests an original monster-catching game instead.
• Space Jam Website Recreation [12:12–12:40]: Recreates the 1996 retro Space Jam website with an interactive toggle switching to a modern 2026 redesign.
• Long-Document Processing & Context Test [12:41–13:41]: Uploads Leo Tolstoy's War and Peace (full text file) to Claude, examines context window limits (200k in web chat vs. 1M API), and produces a visual story breakdown across five movements.
• YouTube Title Generation [13:42–14:25]: Generates non-clickbait YouTube title ideas by providing the Anthropic announcement URL directly in the prompt.

Claims & numbers

• Model details & pricing: The presenter states Claude Opus 4.7 pricing via API remains the same as Opus 4.6 at $5 per million input tokens and $25 per million output tokens [01:48].
• Context windows: The presenter reports that on the Claude website, Opus 4.7 has a 200,000-token context window on standard paid plans (with 500k on Enterprise), while the API supports a 1-million-token context window [01:48, 02:58].
• Effort controls: The presenter notes the API introduces an xhigh ("extra high") effort reasoning setting alongside low, medium, and high, while the web UI provides an automatic "Adaptive thinking" toggle [02:37, 03:07].
• Safety withholding: The presenter states Anthropic withheld the higher-performing "Claude Mythos Preview" due to cybersecurity concerns under Project Glasswing, sharing it only with ~40 selected partner organizations [01:03, 01:40].
• Code generation benchmark: The presenter displays SWE-bench multilingual and multimodal benchmarks showing Opus 4.7 scoring 80.5% compared to Opus 4.6's 77.8% [02:25].

Notable quotes

• "Claude has been and is still the best coding model available today..." [00:13]
• "This is going to choose the level of reasoning based on your prompt." [03:07]
• "I would say that's a pass. That looks fantastic." [10:04]

Assessment
This is a third-party creator review and hands-on feature demo from Skill Leap AI rather than an official Anthropic release. All demonstrations are performed live inside Claude's web interface and artifact rendering sandbox; outputs are genuinely generated, though prompts are deliberately structured to highlight Claude's strong frontend coding abilities.

Described by gemini-3.8-flash on 2026-09-29 from the video's audio and frames.