
Claude Fable 5.2 y Mythos 5.2: qué cambia, cuánto cuesta y nuestras pruebas
Claude Fable 5.2 and Mythos 5.2: what changes, how much it costs and our tests
Title history
- Claude Fable 5.2 y Mythos 5.2: qué cambia, cuánto cuesta y nuestras pruebas — observed on 20261111-1439 UTC (YouTube)
Keywords
Summary
154 words
Critical Evaluation
Value of the Information & Strength of the Argument
The discussion provides practical insights from direct usage, offering a mixed perspective on the new models’ strengths and weaknesses. The hosts argument that while the new models perform exceptionally well on certain benchmarks and creative tasks, the cost and occasional dead ends make the previous Opus 5.5 more suitable for their daily workflow. They substantiate their opinions with specific test results and community examples, but some claims lack rigorous backing, making the overall argument persuasive yet not definitive.
Scientific Rigor, Source Quality, Title Accuracy
The hosts refer to the official Anthropic announcement and cite specific metrics like pricing and performance improvements. They also reference community achievements without fully verifying them, which somewhat weakens the rigor. The title matches the content, but the video includes many personal opinions and subjective assessments rather than a systematic comparison. No comments were provided for analysis.
150 words
Title / Content Match
The title accurately reflects the content: it covers the new models, their cost, and the hosts' practical tests.
Quality & Reliability
6/10
The video provides hands-on impressions and comparisons, but relies on anecdotal evidence and lacks systematic methodology. Some claims are based on community examples without independent verification.
Key Moments
Markers derived by PSI from the transcript: the creator did not define chapters.
- Introduction of Fable 5.2 and Mythos 5.2, including the fact that Fable built an operating system in 8 days.
- Official announcement details: pricing ($8 per million input tokens, $40 per million output), speed improvement, and safety differences between Fable and Mythos.
- Discussion of security tests and an anecdote about Fable identifying a vulnerability but refusing to fix it.
- Community achievements: 1000 agents building OS, verification of Fermat's Last Theorem in 30 hours, and hardware generation experiments.
- Voice test discrepancy: the model identifies as Fable 5.1 in conversation, not 5.2.
- First test: Bart Simpson ASCII art fails due to copyright concerns, resulting in a generic boy with a hat.
- Alchemy test: praised as the best so far, with smooth drag-and-drop, ingredient recommendations, and UX enhancements.
- Animation test: vector-based generated video with excellent quality, the best they've seen.
- Discussion of cost vs. benefit, and why they prefer Opus 5.5 over the new models.
- Final reflections on model overcomplication and the potential risks of extrapolating unexpected features.
Cited Sources
- Claude Fable and Mythos 5.2 announcement — Official Anthropic release notes with details on the models' capabilities and safety.
- Full episode of El Test de Turing (starting point around 1h:00) — Reference to the full episode where this segment originates.
Concurring Sources
- Claude Fable and Mythos 5.2 announcement — The official source supports the claims about model capabilities and pricing discussed in the video.
Contribution & Novelties
This video offers practical, hands-on testing of Claude Fable 5.2 and Mythos 5.2, providing a user perspective that complements official marketing materials. It highlights specific strengths (e.g., alchemy and animation generation) and weaknesses (e.g., refusal on copyright issues) that may not be apparent from benchmark scores. The discussion on cost-effectiveness and the decision to stay with an older model adds value for potential adopters.
Pour aller plus loin :
- Large language model - Wikipedia — Provides background on the technology behind these models.
- AI alignment - Wikipedia — Relevant to the safety differences between Fable and Mythos.
- Anthropic research page — Official research publications on Claude and related safety topics.
110 words
Radar Profile
The score profile indicates a moderately informative video with above-average technical depth but only average reliability and quality. The balance between information quantity and potential bias suggests a subjective review rather than a rigorous analysis.