Detailed Analysis
Anthropic has launched Claude Fable 5, a publicly accessible model the company describes as "Mythos class" — a new capability tier sitting above Opus in the company's model hierarchy. Simultaneously, Claude Mythos 5 has been released, but exclusively through Project Glasswing, a restricted program operated in collaboration with the U.S. government focused on cyber defense and critical infrastructure. Fable 5 carries pricing of $10 per million input tokens and $50 per million output tokens, representing double the cost of Opus, while Mythos 5 previously carried pricing five times that of Opus during its preview phase — a differential that signals Fable 5 represents a deliberately constrained version of the full Mythos capability set. The launch confirms that Mythos-level capabilities have not been fully democratized; rather, Anthropic has released a version with cyber safeguards applied, making it suitable for general deployment while preserving the most powerful iteration for a narrowly vetted user base.
The access structure for Fable 5 introduces notable commercial complexity. From launch through June 22nd, the model is bundled into Pro Max, Team, and seat-based Enterprise subscription plans at no additional charge. Beginning June 23rd, however, Anthropic will shift Fable 5 to a usage-credit model billed on token consumption, with the stated intention of restoring it to standard subscription plans once compute capacity allows. This temporary inclusion-then-removal approach reflects the acute tension Anthropic faces as a company that has recently filed to go public while operating at a loss, subsidizing model access at rates well below actual compute costs. The decision to let users freely experiment with Fable 5 during a defined two-week window appears designed to accelerate adoption and benchmark real-world usage patterns ahead of the transition to metered billing.
The distinction between Fable 5 and Mythos 5 is central to understanding Anthropic's tiered deployment philosophy. Mythos 5, available only through Project Glasswing, is explicitly described as Fable 5 with cyber safeguards lifted — meaning the publicly accessible model has had capabilities in sensitive domains such as cybersecurity deliberately restricted. Users who accessed the earlier Claude Mythos preview through Glasswing are eligible to upgrade to Mythos 5. This architecture reflects Anthropic's broader safety posture: rather than withholding a model entirely until it clears all safety thresholds, the company releases a capability-limited version publicly while continuing to deploy the full model in controlled, vetted environments. The blog post's framing of Fable 5 as a "Mythos class model that we've made safe for general use" is a direct acknowledgment of this bifurcation.
The launch connects to a broader pattern in frontier AI development in which the most capable models are not uniformly accessible. Anthropic's Project Glasswing mirrors analogous arrangements at other labs where government partnerships and national security considerations shape which capabilities reach the open market and which remain gated. The company's decision to benchmark Fable 5 as "state-of-the-art on nearly all tested benchmarks" across software engineering, scientific research, and vision tasks positions it as a direct competitive response to models from OpenAI and Google, while the Mythos 5 track suggests Anthropic is building a separate commercial and governmental lane for its most capable systems. As Anthropic moves toward a public offering, this dual-track strategy — broad consumer access to a safety-tuned frontier model alongside restricted access to a more capable government-facing version — will likely define the company's near-term product and revenue architecture.
Read original article →