Claude Opus 4.8 Is HERE – Is THIS the Best Model Yet?

Summary Bijan Bowen reviews and benchmarks Anthropic’s newly released frontier model, Claude Opus 4.8. Across desktop, Cowork, Claude Code, and web interfaces, he puts the model through a battery of complex coding and generation tests—including browser operating systems, 3D games, animated marketing SVGs, and 3D simulations—comparing its outputs against Claude Opus 4.7 and GPT-5.5.

What is shown

• 00:10 — Review of Anthropic’s “Introducing Claude Opus 4.8” blog post, detailing benchmark scores, dynamic workflows, fast mode, and safety evaluations.
• 04:32 — Test 1: Browser OS (“NeonOS 1.0”): Opus 4.8 generates an in-browser operating system featuring synthwave styling, live shaders, Spotlight search, notepad, paint app, and two playable 3D games (“Auto City” and “Orbital”).
• 10:14 — Test 2: Animated Marketing SVG: Using Claude Desktop’s Cowork feature, Opus 4.8 produces a synchronized, 60-second animated vector presentation with custom branding for sponsor Oxylabs.
• 12:52 — Test 3: 3D Subway Station FPS (“Line 6”): Generation of a dark subway station environment with lighting controls, moving AI enemies, and first-person shooter mechanics.
• 15:28 — Test 4: 3D Skateboarding Game: Claude Code compiles a standalone C++/OpenGL 3D skateboarding simulation (“Boardwalk Bladerz ‘98”) with tricks, grinding, pedestrian NPCs, and boardwalk environment.
• 18:40 — Test 5: Seinfeld Apartment 3D & Beat ‘Em Up (“Apartment Brawl”): A 3D recreation of Jerry Seinfeld’s apartment turned into a low-poly multi-wave fighting game.
• 22:23 — Test 6: 3D Flight Combat Simulator (“Ace Dominion” / “Sky Strike”): Creation of an aerial dogfighting game with plane selection, projectile tracers, and ground collision effects.
• 24:15 — Test 7: Frontend Landing Page (“Ravioli Rosso”): Interactive, CSS/JS food brand page with floating interactive SVG elements and dynamic sliders.
• 25:23 — Test 8: 3D Printer Simulation: A Three.js simulation of an FDM 3D printer laying filament toolpaths to print squares, circles, and triangles.
• 29:46 — Test 9: 3D Arcade Machine (“Omni Racer”): An end-to-end task turning a photo of a physical arcade steering wheel cabinet and an unaligned car sprite sheet into a full 3D arcade cabinet running a playable 3D racer on its virtual screen.
• 37:17 — Test 10: Drum Kit Simulation (“Drum Kit Designer”): A playable 3D drum kit running Web Audio API synthesis with interactive kits and four automated genre backing tracks (Rock, Funk, Hip Hop, Jazz).

Claims & numbers

• Release Date: Claude Opus 4.8 launched on May 28, 2026.
• Pricing: Retains Opus 4.7 pricing at $5 per million input tokens and $25 per million output tokens; Fast mode is priced at $10 per million input and $50 per million output (working at 2.5x speed).
• Stated Benchmark Figures:
  • Agentic coding: 69.2% (vs. Opus 4.7 at 54.2%, GPT-5.5 at 58.6%).
  • Agentic terminal coding: 74.6% (vs. Opus 4.7 at 66.1%, GPT-5.5 at 78.2%, Gemini 3.1 Pro at 70.3%).
  • Multidisciplinary reasoning: 49.8% (vs. 46.1% for Opus 4.7).
  • Agentic computer use: 83.4% (vs. 82.0% for Opus 4.7, 78.7% for GPT-5.5).
  • Knowledge work: 1,890 (vs. 1,753 for Opus 4.7, 1,769 for GPT-5.5).
  • Agentic financial analysis: 53.9% (vs. 51.5% for Opus 4.7).
• The presenter notes Anthropic mentions upcoming “Mythos-class” models from Project Glasswing with higher intelligence than Opus.

Notable quotes

• 15:57: “I would say, this is rather frustrating. Extremely so.”
• 28:03: “It looks like a Hershey’s Kiss!”
• 36:05: “This is deeply, deeply impressive. And this is exactly what I wanted.”

Assessment This is an authentic, hands-on independent review and technical stress-test of Claude Opus 4.8 by a community developer. The demonstrations run live in desktop software, terminal, and browsers, frankly highlighting both glitches/hangs and exceptionally strong end-to-end multi-asset 3D generation capabilities.

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