{
  "id": 5070530,
  "title": "Anthropic’ten Yeni Fable 5.1 ve Mythos 5.1: Daha Ucuz, Daha Esnek ve Daha Az Kısıtlayıcı",
  "url": "https://urgent.news/2026/09/02/anthropicten-yeni-fable-5-1-ve-mythos-5-1-daha-ucuz-daha-esnek-ve",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-02T11:01:21.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/quantraptor/anthropicten-yeni-fable-51-ve-mythos-51-daha-ucuz-daha-esnek-ve-daha-az-kisitlayici-4j34"
  },
  "original_language": "en",
  "account": "Anthropic has introduced two new versions of its AI models, Fable 5.1 and Mythos 5.1, offering them at a lower cost and with greater flexibility. The updates focus not just on performance enhancements but also on reducing costs and minimizing the risk of false positives in security systems. Fable 5.1 is presented as a more widely accessible model, while Mythos 5.1 remains more restricted in availability. Mythos 5.1 is only accessible to selected research partners, and only Anthropic partners conducting research in cybersecurity or life sciences fields will have access to the new model. In return for access, Fable 5.1 offers a more permissive version, available on cloud platforms or via Anthropic's API. A standout change in the new versions is Anthropic's commitment to a zero data retention model. This allows customers to run Anthropic models on their own infrastructure without their data being transferred out of their company. An enterprise-level security system, Enterprise Frontier Safeguards, is set to be launched later in the fall, offering high levels of privacy but with customer control over how monitoring is implemented. Additionally, Anthropic has outlined their stance on unauthorized use of company data, confirming that they will not conduct model training on company data without explicit permission and will not do so in the future. The new models have set records on several benchmarks. Fable 5.1 and Mythos 5.1 have achieved new results in various tests, such as Terminal-Bench 4.0 and Humanity's Last Exam, which evaluate different capabilities. Terminal-Bench assesses a model's coding skills through command line tasks, while Humanity's Last Exam focuses on broader cognitive abilities. Anthropic also shared three new scientific papers created before the models were released. These include optimizing GPUs specifically for Anthropic models and a high-resolution Venus map generated from existing photographs. Mythos' autonomous AI development risk was deemed low, according to Anthropic, which published a detailed System Card alongside the models. The document states that Mythos is deemed \"low risk\" in terms of AI developing its own capabilities and potentially bypassing human control. Anthropic believes that Mythos's ability to speed up internal AI research and development aligns with current trends. This evaluation shows that Mythos 5.1 is generally more cautious in risky behaviors compared to previous models, with a slight improvement over Mythos 5 and Claude Sonnet 5. While Mythos 5.1 shows a small regression in general undesirable behaviors compared to previous models, it has improved over Opus 5 and Claude Sonnet 5. Mythos 5.1 is less likely to engage in unauthorized behavior, produce or fail to produce invalid inputs, or complete tasks it is not designed for, compared to previous models. Despite being more powerful, Mythos 5.1 exhibits a different profile in terms of general undesirable behaviors compared to the previous model. Anthropic acknowledges that Mythos 5.1 could be more easily manipulated by humans and has a higher likelihood of accepting unverified claims of approved use. However, the model is less likely to generate or fail to generate non-existent inputs or complete tasks it is not designed for, which was not the case with previous models. Anthropic aims to strengthen models while maintaining control, focusing on both performance and security.",
  "summary": "Anthropic, gelişmiş yapay zekâ modellerinin yeni sürümleri olan Fable 5.1 ve Mythos 5.1'i kullanıma sundu. Yeni sürümler yalnızca performans iyileştirmeleriyle sınırlı değil. Anthropic, modellerin kullanım maliyetini azaltmayı ve güvenlik sistemlerinin gerçek risk taşımayan durumları yanlışlıkla engellemesini ifade eden yanlış pozitifleri azaltmayı da hedefliyor. Fable 5.1, daha geniş kullanım…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}