Tech • IA • Crypto
Anthropic Just Dropped Claude Mythos and Fable 5 (Full Breakdown)
10/10TL;DR
Anthropic has released Claude Fable 5 to the public and expanded access to Claude Mythos 5, introducing major performance gains alongside strict safety controls and higher pricing.
KEY POINTS
Public launch of Fable 5
Claude Fable 5 is now available as Anthropic’s most capable publicly accessible model. It is positioned as a “mythos-class” system adapted for general use, meaning it derives from a more powerful internal model but includes added safeguards. The company claims it surpasses all prior public models across benchmarks including reasoning, coding, and vision.
Mythos 5 remains restricted
Claude Mythos 5 uses the same underlying system as Fable 5 but with fewer safety constraints. Access is currently limited to select cybersecurity organizations and infrastructure providers, with deployment tied to Project Glass Wing in collaboration with the U.S. government. Broader access is expected to expand تدريجياً under controlled conditions.
Security risks drive safeguards
Anthropic highlights significant risks tied to advanced capabilities, especially in cybersecurity and biological domains. The models can identify and exploit software vulnerabilities, potentially lowering the cost and difficulty of cyberattacks. As a result, Fable 5 frequently defers sensitive queries to less capable models such as Claude Opus 4.8, limiting misuse but also reducing usability in flagged areas.
Performance leap in coding
Benchmark results show a notable jump in agentic coding, with Fable 5 and Mythos 5 scoring 80.3%, compared to 69.2% for Opus 4.8. Early enterprise testing suggests substantial productivity gains, with Stripe reporting that tasks previously requiring months of engineering work were completed in days using Fable 5.
Advances in vision and multimodal tasks
The model introduces major improvements in vision capabilities, including extracting precise data from complex scientific visuals and reconstructing web application code directly from screenshots. These advances are seen as critical for future AI systems that interact with computer interfaces in real time.
Efficiency across effort levels
Anthropic claims Fable 5 outperforms competing frontier models even at medium compute settings, offering a balance between cost and performance. Users can scale effort levels from low to maximum, with higher settings delivering stronger results at increased token usage.
Pricing doubles over previous models
Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens, roughly double the cost of Opus 4.8. Despite the increase, the pricing is lower than some expectations, as early speculation suggested significantly higher costs for mythos-class systems.
Breakthroughs in scientific research
Internal testing indicates strong potential in drug discovery and protein design, where Mythos 5 reportedly accelerated parts of the research process by up to 10×. In some cases, the model matched or exceeded skilled human scientists in executing complex bioinformatics workflows without assistance.
CONCLUSION
Anthropic’s latest release signals a significant leap in AI capability, but its impact will depend on how effectively safety restrictions, pricing, and controlled access balance innovation with real-world usability.