01What Claude Opus 5 Is and How It Works

Claude Opus 5 is Anthropic's newest AI model. It is a thoughtful and proactive model that comes close to frontier intelligence at half the price of Fable 5. In this episode, Julian Goldie tests it across 50 tasks on GoldieBench and shows how it performs on various benchmarks.

Claude Opus 5 works by using advanced reasoning and agentic capabilities. It can handle complex tasks like coding, knowledge work, and computer use. It is available today on Claude.ai and inside the agent operating system that Julian uses.

02Main Features and New Updates

Claude Opus 5 has several standout features. It excels at agentic terminal coding, outperforming Fable 5 and GPT 5.6 Sol on FrontierBench. It is also better at knowledge work, problem solving, and agentic search.

On ARC AGI 3, where AI models solve novel problems, Opus 5's score is three times as high as the next best model.

There are two new updates: mid-conversation tool changes on the Claude platform, and automatic fallbacks on the API. The API fallback routes to another model when it doesn't like what you're doing, which Julian warns may cause complaints.

  • Agentic terminal coding
  • Knowledge work
  • Problem solving
  • Agentic search
  • Computer use
  • Multidisciplinary reasoning

03How to Set Up and Get Started with Claude Opus 5

Getting started with Claude Opus 5 is simple. It is available today on Claude.ai, so you can access it directly. Inside the agent operating system, Julian has already plugged in Opus 5, and it was super easy to set up.

  1. Go to Claude.ai and select Opus 5.
  2. If you have the agent operating system, open the workspace and add Opus 5 as one of your agents.
  3. It goes straight into the memory system, so you don't have to re-prompt it.
  4. You can also plug Claude Code into other models like GLM 5.2, Kimi K3, Quen 3.8, or GPT 5.6 if you prefer.

Julian also mentions that with Opus 5, you can use 100% of your Claude subscription usage, unlike Fable 5 which only includes 50%.

04How to Use It: Examples from the Episode

Julian shows six one-shot demos built with Claude Opus 5. The Dragon Realm game looks like a Skyrim-style RPG with nice graphics, ambience, and no visible bugs. The snow drop and northern lights background are super nice.

The Doom Ray Maze runs without lag, and controls feel good. The Neon Snake game is a 3D snake game that is fun to use. There is also an air combat game that is a bit hard to control due to physics, but the quality and vibe are good.

One demo didn't get past the screen, so it's not perfect every time. Julian emphasizes that you should test it yourself and not believe the benchmarks blindly.

  • Use loops: give Opus 5 tasks and let it iterate until it succeeds.
  • Use it for software engineering tasks, which perform well on FrontierBench and Cursor Bench.
  • Use fast mode for 2.5 times the default speed, but note it costs twice the base price.

05Results and Comparisons vs Other Models

On FrontierBench, Claude Opus 5 scores 43.3% versus Fable 5 at 33.7% and GPT 5.6 Sol at 34.4%. On ARC AGI 3, Opus 5's score is three times higher than the next best model. On Zapier automation bench, Opus 5's pass rate is around 1.5 times the next best model for the same cost per task, even at its lowest effort setting.

On OS World, a computer use benchmark, Opus 5 outperforms every other model at any given cost, surpassing Fable 5's best result at just over a third of the cost.

On cost per task, Opus 5 is cheaper than Fable 5 and Opus 4.8 even when effort levels go up. GPT 5.6 Sol is still slightly higher. Julian notes that Opus 5 is stronger than Opus 4.8 outside of security tasks but remains substantially behind Miphos 5 at developing exploits.

06Tips, Limits, and Mistakes to Avoid

Julian advises that you should always test Claude Opus 5 yourself. Benchmarks can be misleading, and not every task will be perfect. In the episode, one demo failed to get past the screen, so don't be disappointed if you see failures.

Be aware of the automatic fallback on the API, which could route to another model and cause complaints.

Pricing: $5 per million input tokens and $25 per million output tokens, same as Opus 4.8. Fast mode runs at 2.5 times the default speed but costs twice the base price. Julian also warns that Fable 5 only includes 50% of your subscription chat, while Opus 5 allows 100%.

  • Test it yourself; don't rely solely on benchmarks.
  • Be cautious with automatic fallbacks on the API.
  • Use fast mode only if you need speed and accept the higher cost.

07Who It Is For and Why It Matters

Claude Opus 5 is for anyone who needs frontier-level AI without the high cost. It matters because Anthropic is under pressure from competitors like GPT 5.6, Grok 4.5, and Chinese models like Quen 3.8 Max and Kimi K3. Opus 5 offers affordability and availability, making it a strong choice for developers, businesses, and AI enthusiasts.

Julian mentions that it is already integrated into the agent operating system, which works with Hermes, Anti-gravity, and Claude. It supports tools like Open Design, Open Montage, and video editors. This makes it useful for building, automating workflows, and saving time in various projects.

08Pricing, Availability, and Security Notes

Claude Opus 5 is priced at $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8. It is available today on Claude.ai. You can also use it inside the agent operating system, which Julian offers through the AI Profit community.

Security-wise, Opus 5 has safeguards designed to allow developers to identify and fix software vulnerabilities while blocking high-risk uses. It is stronger than Opus 4.8 outside of security tasks but remains behind Miphos 5 at developing exploits.