Claude Opus 5 Is Blind. It Still Built a AAA Game.

Summary

Josh Kale and Ejaaz Ahamadeen review Anthropic's Claude Opus 5, a new AI model that excels in visual intelligence and agentic sub-tasks at half the cost of Fable 5. They showcase one-shot game demos, discuss its record-breaking ARC AGI 3 score, and compare its safety behavior and token efficiency to Fable 5 and open-source alternatives. The hosts conclude that Opus 5 is the best everyday model for most users.

• Claude Opus 5 offers high-quality visual and agentic capabilities at 50% of Fable 5's cost.
• Live demos include a structurally accurate Brooklyn Bridge rendering and a first-person shooter built from a single prompt.
• The model achieved a 30% score on ARC AGI 3, four times higher than the previous best frontier model.
• Opus 5 triggers safety refusals 85% less often than Fable 5, making it more practical for daily use.
• Agents and sub-agents can be orchestrated to build complex projects autonomously.
• Anthropic's closed safety alignment contrasts with the open-source approach of models like Kimi K3.
• NVIDIA and Microsoft signed an open-source waiver, but Anthropic has not, citing safety reasons.
• The hosts recommend Opus 5 as the default model for most general-purpose tasks over Fable 5 or GPT-5.6 Sol.