Matthew Berman

Fable 5.1 is the BEST AI Model (Cancel Your Other Subscriptions)

Sep 1, 2026 19 min
anthropicfable 5.1ai benchmarksdata privacyfrontier models
Watch on YouTube Follow Matthew Berman on Rundown — free

Summary

AI summaries can be incomplete or wrong. Verify anything important against the original video.

This video reviews the release of Anthropic's new Fable 5.1 and Mythos 5.1 models, analyzing their cost-to-performance improvements, safety features, and benchmarking results.

The video examines the release of Anthropic's new frontier AI models, Fable 5.1 and Mythos 5.1. The creator highlights significant improvements in performance-per-dollar, particularly through reduced costs on cache reads. He conducts several benchmarks, including 'Terminal-Bench' and 'Humanity's Last Exam', demonstrating how these models often surpass predecessors and competitors. A substantial portion of the review discusses the new 'Enterprise Frontier Safeguards' (EFS) and 'zero data retention' policy, which aim to address corporate concerns about data privacy. Furthermore, the video covers safety upgrades, specifically aimed at preventing 'distillation attacks' and compliance with EU AI Act watermarking requirements. The review concludes by testing these models' creative capabilities, finding them competitive in tasks like web generation, though noting they still have limitations, such as hallucinated specifics in website content.

Verdict

Claude Fable 5.1
ai model

Fable 5.1 is the best frontier model available, offering significant performance gains and cost reductions, especially for agentic tasks, despite not being the absolute cheapest in raw token cost.

Buy

Pros

  • ✓ Significant performance increase per dollar 1:38
  • ✓ Improved agentic capabilities and coding tasks 5:39
  • ✓ Enhanced enterprise data privacy (zero data retention) 4:09

Cons

  • ✗ Higher token usage (1.7x output tokens) compared to competitors 13:27
  • ✗ Still more expensive per task than some smaller models 13:20
  • ✗ Potential for hallucinatory data generation in web artifacts 17:10

Specs

Cache read cost $0.25 per million tokens 12:32

Compared to

  • GPT-5.6 Sol

    Competitive in performance but often beats it on price for certain tasks.

  • Claude Opus 5

    Opus 5 was excellent, but Fable 5.1 offers better reasoning and agentic performance.

Best for

  • Software developers
  • Enterprise companies requiring data privacy
  • Research labs

Not for

  • Budget-constrained personal users
  • Users looking for raw cheapest tokens

Key Points

  • 0:18 Introduction of Fable 5.1 and Mythos 5.1 as new frontier models.
  • 1:38 Detailed cost analysis: 25% cheaper for typical workloads, up to 45% for high agentic workloads due to cache reads.
  • 4:09 Discussion on Enterprise Frontier Safeguards (EFS) and data retention concerns.
  • 5:39 Benchmark results: Fable 5.1 significantly outperforming on 'Terminal-Bench Science 0.1'.
  • 11:32 Comparison between Mythos 5.1 and Fable 5.1, noting they are the same model but with different guardrails.
  • 14:45 Hands-on testing: Generating websites and 3D scenes to evaluate output quality.
  • 16:46 Review of the 'distillation attack' prevention safeguard.

Worth watching if: You are a professional or developer trying to decide if Anthropic's new models are cost-effective replacements for your current AI stack, or if you need to understand their new data privacy and safety safeguards.

Get every Matthew Berman video extracted like this

One daily email with structured extracts of every channel you follow. Free tier covers 15 videos a month.

Sign in with Google

No credit card. Free tier forever.

Watch on YouTube