Video summary

People are Creating INSANE Things With Claude Fable 5 (Mythos 5)

Main summary

Key takeaways

Technology

Summary of the video’s technological concepts & product features

Anthropic has released Claude Fable 5 (public version of “Mythos 5”), described as a general-use model with capabilities that had previously been considered too powerful to fully release. The video focuses on how quickly users can generate complex software and media outputs—often from a single prompt—and discusses benchmark claims and pricing/cost-performance.

1) One-prompt game creation (coding at scale)

Minecraft-style clone (HTML + Claude Fable Max)

  • Built by a user named Angel
  • Includes most core Minecraft-like functionality, visuals, and movement feel
  • Uses shaders and is reported to look visually strong
  • Minor issue mentioned: character hand transparency
  • Claimed cost: ~$30 in API usage to generate a complete playable game-like experience

Pokémon Fire Red beaten from screenshots

  • Claude reportedly completes Pokémon Fire Red using only screenshots
  • No help claimed such as coordinates/mini-map, memory hacks, or extra assistance
  • Reported completion time: ~50 minutes

Pokémon clone “one-shot” generator

  • Generates a full Pokémon-style game via a single prompt
  • Took about 1 hour of reasoning
  • Produced ~8,000 lines of code
  • Claims include all 151 Gen 1 Pokémon with real-like data, such as:
    • Front/back sprites
    • Party icons
    • Cries
    • Base stats
    • Types
    • Level-up moves
    • Evolutions
    • Catch rates
    • Growth curves

Call of Duty Zombies clone

  • Created with no plan mode and no back-and-forth
  • Just a single prompt
  • Prompt gist: create a single-player COD zombie clone with good shooting feel
  • Visual fidelity and “feel” described as surprisingly good

2) Websites generated quickly with premium UX

One-shot self-introduction / story website

  • Generated with strong visual design, minimal modern typography, and high-quality image effects
  • Noted “premium” interaction, including cursor/page behavior
  • Estimated manual development cost: ~$10,000, but generated via one prompt

An additional Claude-generated website is also shown, described as similarly impressive.

3) 3D worlds inside the browser (e.g., 3.js)

  • A custom browser-based 3D world built using 3.js
  • Runs directly inside a web browser
  • Emphasis: realism is “pretty good,” and it was likely generated in one prompt

4) Motion graphics / audio-visual generation

Music + visualizer created with code

  • Includes generated visual animations synced to melody
  • Described as soothing and technically impressive

Motion graphics videos using “Hyperframes”

  • Claude generates animation code
  • Hyperframes converts code into rendered motion graphics
  • Video quality described as very cool, with only minor caveats expected to be fixable via follow-up prompts
  • Voiceover may be generated with 11 Labs (mentioned as likely used)

5) Model performance claims (benchmarks & “frontier” reasoning)

The video quotes Anthropic’s launch-style claims:

  • Claude Fable 5 is “safe for general use”
  • It “exceeds” previously generally available state-of-the-art on many benchmarks
  • Examples include claims that Stripe compressed engineering time, and strong results on:
    • Cognition’s Frontier code
    • Hebbius finance benchmark

An informal comparison is also made:

  • GPT-5 is framed as much weaker than Claude Fable 5

A more concrete reasoning comparison is referenced:

  • Opus 4.8 scores ~13.4%
  • Claude Fable 5 scores ~30%
  • On a “Frontier Code Diamond” benchmark described as a new gold standard

6) Pricing / cost-effectiveness analysis

Availability and pricing model

  • Currently available on the Claude plan until June 12
  • After June 12: shifts to usage-based credits (pay based on usage)

Cost comparisons emphasized

Claude Fable vs Claude Opus 4.1

  • Opus 4.1: $15/million input tokens, $75/million output tokens
  • Fable: $10 input, $50 output

Claude Fable vs GPT 5.5 Pro

  • GPT 5.5 Pro: $30 input, $180 output
  • Fable: $10 input, $50 output
  • Claim: ~70% cheaper than GPT 5.5 Pro, while also described as “at least a few times better”

Overall conclusion

  • Even if some users consider it expensive, the video argues the intelligence gained makes it cost-justified.

7) “Everything created by the model” premise

The video includes a claim that at least one showcased video was created by the AI with no human intervention beyond the initial prompt, emphasizing end-to-end generation speed.

Main speakers / sources mentioned

  • “Angel” — user who created the Minecraft HTML clone example
  • Anthropic (launch announcement quoted) — source of benchmark/capability and “safe for general use” claims
  • The video narrator/creator — speaks throughout and frames the examples and price analysis

Original video