September 3, 2026, (Inside AI) — Anthropic has released two new AI models, Claude Fable 5.1 and Claude Mythos 5.1, targeting coding, research, and specialized scientific work. The models share a common architecture but carry different safety guardrails and access levels.
Fable 5.1 is generally available across Anthropic's platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. Mythos 5.1 is restricted to a small set of vetted users and organizations in cybersecurity and life sciences.
The launch marks a continued push by Anthropic to balance frontier capability with controlled deployment. Fable 5.1 improves long-running task performance and cuts typical workload costs by about 25 percent compared to Fable 5. Highly agentic workloads could cost up to 45 percent less.
Cache read prices dropped 75 percent to $0.25 per million tokens. Regular input and output prices remain at $10 and $50 per million tokens, respectively.
Anthropic's internal benchmarks show Fable 5.1 scoring 52.6 percent on Terminal-Bench-Science, up from 24.7 percent for Fable 5 and 29 percent for Opus 5. On CursorBench, it reached 73.4 percent versus 70.5 percent and 70 percent for the earlier models.
Real-world examples include investment firm Millennium using Fable 5.1 to diagnose a rare software crash that had stumped engineers for years. The model traced the bug to an external vendor library.
Mythos 5.1 showed strength in protein design. Across 12 targets, nearly 50 percent of its designs were confirmed as viable binders, far above typical hit rates of 10 to 15 percent.
Fable 5.1 also contributed to planetary science. It trained a neural network to build a higher-resolution elevation map covering about one-third of Venus from NASA's Magellan mission data. The new map reveals details at two to three kilometers, compared with 10 to 20 kilometers previously.
Safety Tuning Separates the Two Models
Anthropic adjusted safety systems to reduce friction. Updated cybersecurity safeguards produce about 60 percent fewer interventions per session for Claude Code users compared to Fable 5. The model can identify software vulnerabilities for defensive work, but exploit generation and penetration testing remain limited.
Mythos 5.1 is available only through trusted access programs. Anthropic plans to expand access later. Fable 5.1 is live now on major cloud platforms.
The Fable and Mythos line began with Claude Mythos Preview in April. It was offered to a limited group of cyber defenders through Project Glasswing. That preview raised concerns about AI accelerating vulnerability discovery. Anthropic found it could autonomously detect unknown flaws in major operating systems and browsers, sometimes turning them into working exploits.
Attackers could use similar systems to find and exploit vulnerabilities faster and with less expertise. Anthropic warned that the gap between detection and exploitation could shrink, leaving less time to patch. That concern drove the decision not to release Mythos Preview publicly.
Fable 5 and Mythos 5 followed in June, with Fable as the safeguarded general-use version and Mythos for trusted partners. The 5.1 release refines that approach with better capabilities and more precise safeguards to cut false positives.
The dual-track strategy reflects a broader industry debate over how to release powerful models without enabling misuse. Anthropic's approach contrasts with open-weight releases from some competitors, but the company has not indicated any shift toward fully open access for its most capable systems.
