EN

Opus 5.5 Is Crazy Good and GPT-6-Sol Launched Too

#AI Tools
2026-09-23 14:2113 min read

Content Introduction
Ask Questions

The video is a rapid first-look review of two major AI model launches on the same day: Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6 Soul/Luna. The speaker argues Opus 5.5 is the standout release, describing it as faster, cheaper, and smarter than Anthropic’s prior flagship, with strong benchmark gains and impressive real-world demos in coding, automation, graphics, games, and animation. OpenAI’s new models are presented as useful but less dramatic improvements, mostly notable for lower pricing and broader availability. Overall, the video emphasizes how quickly AI capabilities are advancing, especially in creative and agentic tasks, while teasing a fuller weekly roundup to come.

Key Information

  • The speaker is reviewing major AI model announcements made on the same day while at Meta Connect in Palo Alto.
  • Anthropic released Claude Opus 5.5, which is described as better than previous Claude models, faster, and cheaper.
  • Claude Opus 5.5 reportedly matches or beats many benchmarks, including agent programming, coding, and general intelligence metrics.
  • The speaker emphasizes that Opus 5.5 is now available widely and that its cost per task is lower despite using more tokens.
  • The video highlights impressive demos created with Opus 5.5, including animations, a Game Boy emulator, games, and Blender-based work.
  • OpenAI also released GPT-6 Soul and GPT-6 Luna, but these are portrayed as less impressive than Opus 5.5.
  • GPT-6 Soul and Luna are said to be cheaper and faster than prior versions, but not as strong as Anthropic's new model.
  • The speaker compares benchmark rankings, noting Opus 5.5 leading on some aggregate analyses while GPT-6 Soul ranks lower.
  • A key theme is that real-world output, demos, and cost-per-task matter more than raw benchmark scores.
  • The speaker concludes that Opus 5.5 is the main news of the day, with more AI announcements and demos expected later in the week.

Timeline Analysis

Content Keywords

Claude Opus 5.5

Anthropic released Claude Opus 5.5, presented as its new top model. The speaker says it is better than Claude 5.1, cheaper than Opus 5, and faster, with strong performance across agent programming, coding, computer use, visual recognition, and intelligence benchmarks. It is also available in Claude for paid users.

GPT-6 Soul and GPT-6 Luna

OpenAI released GPT-6 Soul and GPT-6 Luna shortly after Anthropic's announcement. These models are cheaper and faster than previous versions, but the speaker считает them less impressive than Claude Opus 5.5 and still behind Anthropic's strongest model in overall performance.

AI model benchmarks and cost efficiency

The video compares benchmark results and pricing across models, including Artificial Analysis, BusyBench, Terminal Bench, and other tests. Claude Opus 5.5 leads in intelligence scores, while GPT-6 Soul shows improved price-to-performance but weaker overall results. The main theme is that newer models are becoming much more cost-efficient.

AI-generated animations and games

The speaker highlights impressive demos created with Claude Opus 5.5, including JavaScript animations, a Game Boy emulator, a snake game, a flight simulator, and a Dark Souls-style project. These examples are used to show how far AI-generated graphics, games, and animation have advanced.

Meta Connect and weekly AI news roundup

The speaker is reporting from Palo Alto during Meta Connect and mentions that more announcements may come from Meta and YouTube’s Made on YouTube event. He says he will cover Opus 5.5, GPT-6 Soul, Jev, and other AI news in his Friday roundup video.

Related questions&answers

What new AI models were announced in the video?

The video discusses two major releases: Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Soul and GPT-6 Luna.

Which model does the speaker consider the biggest news of the day?

The speaker says Claude Opus 5.5 is the main news and the most impressive release.

How does Claude Opus 5.5 compare to previous Anthropic models?

It is described as better than Claude 5.1 and cheaper than Opus 5, while also being faster.

What performance claims are made about Claude Opus 5.5?

Anthropic claims it performs at the level of Claude 5.1 on most tasks, while costing less.

How much cheaper is Claude Opus 5.5 compared to Opus 5?

The speaker says it costs 40% less than Opus 5.

What are the pricing changes for Claude Opus 5.5?

Input tokens are said to cost $4 per million and initial tokens $20, down from $5 and $25 previously.

Why does the speaker say the overall task cost is still not dramatically lower for Opus 5.5?

Because although token prices dropped, Opus 5.5 uses significantly more tokens per task.

What benchmark does the speaker reference for evaluating AI models?

The speaker mentions benchmarks like Agent Programming, Terminal Bench, Automation Bench, Humanities exam, Artificial Analysis, and others.

What score does Claude Opus 5.5 get on Artificial Analysis?

The speaker says Claude Opus 5.5 leads with a score of 58.

How does Claude Opus 5.5 compare to Fable 5.1 and GPT-6 Astra in Artificial Analysis?

The speaker says Fable 5.1 and GPT-6 Astra were previously tied at 53, while Opus 5.5 now leads with 58.

What is the speaker's view on the importance of benchmarks?

The speaker says benchmarks matter less than actually seeing, testing, and using the models in real life.

What kinds of demos are highlighted for Claude Opus 5.5?

The video highlights AI-generated animations, a Game Boy emulator, a game inspired by the Antikythera mechanism, a flight simulator, a Mario Maker-style demo, and other creative outputs.

What does the speaker find impressive about the animation demos?

The speaker says the animations look amazing and are hard to believe were generated without tools like After Effects.

What game-related demo does the speaker mention first?

A Game Boy test or emulator demo created with Claude Opus 5.5.

What other OpenAI model was released the same day?

OpenAI released GPT-6 Soul and GPT-6 Luna a few hours after the Opus 5.5 announcement.

How does GPT-6 Soul compare to Claude Opus 5.5?

The speaker says GPT-6 Soul is cheaper and faster, but not as impressive as Claude Opus 5.5.

How does GPT-6 Soul compare to OpenAI's previous model?

It is said to be slightly better than the previous generation of Soul models, but not a major leap.

How does GPT-6 Luna compare to GPT-6 Soul?

GPT-6 Luna is described as a lower-tier model, cheaper than Soul, and available to more users.

What are the pricing changes for GPT-6 Soul?

The video says input token cost was reduced from $4 to $2 per million and output token cost from $20 to $10 per million.

What are the pricing changes for GPT-6 Luna?

The speaker says its entry price was halved from 20 cents to 10 cents and its exit price from $1.20 to $0.50.

Which model does the speaker think is best overall?

The speaker personally thinks Claude Opus 5.5 is the most impressive model, though an LLM judge in a leaderboard placed GPT-6 Soul first.

What is the speaker's opinion on the leaderboard rankings?

The speaker says to take leaderboard rankings with skepticism and rely on your own eyes.

Where is Claude Opus 5.5 available?

The speaker says it is available almost everywhere and can be used in Claude now.

Who can use GPT-6 Soul and GPT-6 Luna?

They are available for paid ChatGPT Work and Codex users, as well as Pro, Business, Enterprise, and EDU users; free and Go users can use GPT-6 Luna in the desktop app.

What does the speaker say about early access for GPT-6 Soul and Luna?

The speaker says they do not seem to have been widely distributed for early access yet, since few demonstrations are available.

What does the speaker conclude about the two model launches?

The speaker concludes that Anthropic had the stronger release because Opus 5.5 is faster, cheaper, and better than its previous model, while OpenAI's release was more modest.

More video recommendations

Share to: