Claude Fable 5.1 kullanıma sunuldu: Ajan görevlerinde maliyet yüzde 45’e kadar düşebilecek – Teknoloji Haberleri

Anthropic, the AI company, unveiled its latest model Claude Fable 5.1, along with the security‑focused Claude Mythos 5.1.

Fable 5.1 promises up to 45 % lower cost for long‑running agents. Anthropic estimates typical workloads will be 25 % cheaper; for tasks that heavily reuse cached data, savings could reach ~45 %. The savings arise from a 75 % reduction in cache‑read fees, now $0.25 per million tokens, while the base API rates stay unchanged at $10 per million input tokens and $50 per million output tokens.

On coding, research, document preparation, and extended agent tasks, Fable 5.1 outperforms its predecessor. Test results indicate it better identifies root causes of software bugs instead of just showing symptoms. Early‑access testers report faster execution, lower token consumption, and clearer answers on lengthy tasks. Anthropic cautions that real‑world performance depends on the specific task and configuration.

Fable 5.1 reduces false positives for harmless requests and can now be used defensively to spot vulnerabilities in source code. Penetration testing, exploit development, and binary vulnerability scanning remain blocked or redirected to other models. Anthropic estimates the new security system could lower cyber‑security interventions per session by roughly 60 % for Claude Code users. Mythos 5.1 shares the same foundation but offers more flexible security limits for cybersecurity and life‑science work; it is not generally available and only accessible through Anthropic’s trusted‑access program. Fable 5.1 is offered in Claude’s paid plans and via the Claude API on AWS, GCP, and Azure.