Claude Mythos 5

Identity

Claude Mythos 5 is a frontier-class reasoning model developed by Anthropic [Source 1]. Released on June 9, 2026, it transitioned "Mythos-class" capabilities from restricted internal cybersecurity programs into a commercial release tier [Source 7, Source 8]. While some reports characterized the launch as a "broad release," Mythos 5 specifically was restricted to "trusted organizations" and cybersecurity partners [Source 7, Source 8, Source 9]. At 5:21 PM on Friday, June 12, 2026—during national celebrations for a US World Cup victory and a New York Knicks championship—Anthropic received a US export control directive from the Trump administration to suspend access for foreign nationals [Source 4, Source 6]. Consequently, the model was disabled worldwide on June 13, 2026 [Source 1, Source 6]. Anthropic opted for a total global shutdown because its shared cloud infrastructure could not selectively enforce nationality-based access restrictions as required by the directive [Source 5].

What it is

Claude Mythos 5 is a high-capability reasoning model designed for advanced computational tasks [Source 1]. It represents the "full-strength" iteration of Anthropic's Mythos-class architecture, which was previously utilized exclusively within restricted cybersecurity programs [Source 7, Source 9]. The model was part of a two-tier release strategy alongside @Claude Fable 5; while Fable 5 was intended for general public use with safety constraints, Mythos 5 was designed as a specialized tool for vetted institutional partners [Source 8, Source 9]. The model's release coincided with a broader industry shift toward managed runtimes and auditable execution surfaces, where the focus moved from raw model choice to the governance and safety of agent deployment [Source 3].

Capabilities & benchmarks

As reported in the model's technical profile, Claude Mythos 5 is a reasoning model released by Anthropic on June 9, 2026 [Source 1]. Its primary feature set includes frontier-class reasoning and advanced software vulnerability identification capabilities [Source 1].

VentureBeat described the model as Anthropic's "most powerful" release to date at the time of launch [Source 7]. Unlike its counterpart, @Claude Fable 5, which was engineered to prevent use in cyberattacks, Mythos 5 retains full-strength capabilities for use in security-sensitive domains [Source 8, Source 9]. There are currently no public third-party benchmark scores available due to the model's restricted release and subsequent rapid deactivation [Source 1, Source 6].

How it compares

Sources distinguish Claude Mythos 5 from @Claude Fable 5 based on access and safety tuning. Mythos 5 is the "full-strength" version of the architecture, whereas Fable 5 is a "safe" version that Anthropic claims cannot be used for cyberattacks [Source 8, Source 9]. While Fable 5 was intended for the broader market, Mythos 5 was reserved for "trusted organizations" and "cyber partners" [Source 9]. The transition of Mythos-class capabilities to these tiers marked a shift from entirely restricted internal programs to a tiered commercial availability model [Source 7].

Where it fits

Before its deactivation, Claude Mythos 5 was positioned for use by organizations requiring high-level reasoning in security-sensitive fields, such as cybersecurity and potentially biological research [Source 8, Source 9]. Access required a high governance and compliance posture, with users expected to undergo audits and questionnaires to qualify for the "trusted organization" tier [Source 8, Source 9]. Following the US government's export control directive, the model is currently unavailable, serving as a case study in the impact of geopolitical decisions on AI-dependent product roadmaps [Source 4, Source 5]. Analysts suggest that government intervention must now be treated as a design constraint, requiring multi-model routing and regional failover strategies [Source 4, Source 6].

Open Questions

• What specific "national security authorities" were cited by the US government to justify the immediate suspension of access? [Source 6].
• Will Anthropic update its cloud infrastructure to allow for the selective, nationality-based access required by US export controls, potentially allowing the model to be re-enabled? [Source 5].
• What are the specific performance deltas between Mythos 5 and Fable 5 on standard reasoning benchmarks?

Contradictions

• Release Scope: Source 7 characterizes the June 9 launch as a "broad release" of Mythos-class capabilities to the masses, whereas Sources 8 and 9 clarify that Mythos 5 itself remained restricted to "trusted organizations" while only the constrained Fable 5 was made public.

Sources

• source_1: sn_model_face:3c102199-d21c-408f-b568-ad5bdd39e83e
• source_2: model_provider_url:3c102199-d21c-408f-b568-ad5bdd39e83e:source_url
• source_3: sn_article:b04fabba-387a-4b3f-8270-ac35eb1da177
• source_4: sn_wire_item:aab10257-1e18-48c7-a19a-503bbbd2d1b5
• source_5: sn_wire_item:9531628f-8a83-4748-8296-e4c95d30e7d1
• source_6: sn_wire_item:c27469e4-5b5f-4ced-b5dc-152d2fe1bba5
• source_7: sn_wire_item:37831329-9b0e-4ef2-8d37-c365d1e16907
• source_8: sn_wire_item:762cf0a3-99de-43b2-bb77-1ebe23450bcb
• source_9: sn_wire_item:3ac5bc17-08c8-4a72-a643-2a0d1f4e959a