Claude Mythos Development Timeline

From Claude 2 in 2023 through the accidental Mythos leak in March 2026 — every major milestone in the evolution of Anthropic's model family.

Three Years of Rapid Evolution

The Claude model family has undergone a remarkable transformation since its early days. What began with Claude 2 in mid-2023 has expanded into a four-tier system spanning Haiku, Sonnet, Opus, and now the newly revealed Capybara tier occupied by Claude Mythos. Each release brought measurable improvements in reasoning, coding, instruction following, and safety — but the pace of advancement accelerated dramatically through 2025 and into 2026.

This timeline documents every confirmed model release, the pivotal leak event that exposed Claude Mythos to the world, and the market and industry consequences that followed. All dates and details are drawn from Anthropic system cards, confirmed public statements, and verified reporting from Fortune, Bloomberg, The Information, and other publications.

Understanding this trajectory matters because it reveals Anthropic's release cadence, the deliberate expansion of their tier system, and the increasingly central role that safety considerations play in how and when models reach the public. The gap between Claude Opus 4.6 and Claude Mythos is not merely technical — it represents a strategic inflection point for the company and the broader AI industry.

Claude Model Releases

A chronological record of every major Claude model release, from the original Claude 2 through the current Opus 4.6 generation.

  • July 2023
    Claude 2 Release

    Anthropic launched Claude 2, marking the company's first widely available commercial model. Claude 2 established the foundation for Anthropic's approach to helpful, harmless, and honest AI, and introduced the brand to a broad developer audience through the Claude API and consumer-facing chat interface.

  • March 2024
    Claude 3 — Opus, Sonnet, Haiku Release

    The landmark release that established Anthropic's three-tier model architecture. Claude 3 Opus became the flagship for advanced reasoning, Sonnet balanced performance with cost efficiency, and Haiku offered the fastest and most affordable option. This tiered structure — named after musical and poetic forms — became the organizing principle for all subsequent Claude releases and allowed customers to choose the right trade-off for their use case.

  • June 2024
    Claude Sonnet 3.5

    Anthropic released an upgraded Sonnet that surprised the industry by matching or exceeding the original Claude 3 Opus on many benchmarks, while maintaining the mid-tier cost structure. Sonnet 3.5 became one of the most widely adopted models in the Claude family and demonstrated that the mid-tier could advance independently of the flagship.

  • October 2024
    Claude Haiku 3.5 / Sonnet 3.5 (New)

    A dual release that refreshed both the budget and mid tiers. The new Sonnet 3.5 improved on its predecessor across coding and analysis tasks, while Haiku 3.5 brought meaningful capability gains to the lightweight tier — making capable AI accessible to latency-sensitive and cost-constrained applications at scale.

  • February 2025
    Claude Sonnet 3.7

    Sonnet 3.7 continued the pattern of aggressive mid-tier improvement, delivering stronger reasoning and coding performance. By this point, the Sonnet line had become many developers' default choice for production workloads, reflecting Anthropic's strategy of lifting the entire model family rather than concentrating gains solely at the top.

  • May 2025
    Claude Sonnet 4 / Opus 4 Release

    The Claude 4 generation arrived with both Sonnet and Opus variants, marking a generational leap in capability. Opus 4 became the new flagship with substantially improved reasoning depth, and Sonnet 4 brought those gains to the mid tier. This release solidified Anthropic's position as a leading frontier lab alongside OpenAI and Google DeepMind.

  • September 2025
    Claude Sonnet 4.5

    An iterative upgrade to the mid-tier Sonnet line within the 4.x generation, delivering refinements to instruction following, multi-step reasoning, and code generation quality.

  • October 2025
    Claude Haiku 4.5

    The budget tier received its 4.x generation update, bringing Haiku closer to the capabilities of earlier Sonnet models while retaining its speed and cost advantages — continuing Anthropic's pattern of cascading improvements down the tier stack.

  • November 2025
    Claude Opus 4.5 + MCP Launch Release

    A pivotal month for Anthropic. Claude Opus 4.5 brought the flagship tier to new heights, while the simultaneous launch of the Model Context Protocol (MCP) established a standard for connecting AI models to external tools and data sources. MCP quickly gained traction, reaching approximately 2 million SDK downloads within its first weeks and laying the groundwork for an ecosystem that would grow exponentially.

  • February 5, 2026
    Claude Opus 4.6 Release

    The most recent flagship release before Mythos. Opus 4.6 achieved #1 on Terminal-Bench 2.0 with a 65.4% score, 80.8% on SWE-bench Verified, and set new baselines across multiple academic and coding benchmarks. At the time of release, it was the most capable model Anthropic had ever produced — a position it would hold for less than two months before being surpassed by the leaked Mythos results.

  • February 17, 2026
    Claude Sonnet 4.6

    The mid-tier counterpart to Opus 4.6, released twelve days later. Sonnet 4.6 scored 79.6% on SWE-bench Verified and 59.1% on Terminal-Bench 2.0, offering strong performance at a more accessible price point. Together, Opus 4.6 and Sonnet 4.6 represented the state of the art for the Claude family heading into the Mythos era.

The Mythos Leak: March 26-28, 2026

The existence of Claude Mythos was never intended to be revealed this way. A CMS configuration error at Anthropic exposed approximately 3,000 unpublished assets to the public internet, triggering a cascade of journalistic discovery, market disruption, and industry-wide reassessment of AI capabilities.

  • March 26, 2026
    The Discovery Leak Event

    Fortune journalist Bea Nolan discovered a publicly accessible Anthropic data store containing roughly 3,000 unpublished assets — including images, PDFs, audio files, and critically, draft blog posts describing a new model called Claude Mythos. Security researchers Roy Paz of LayerX Security and Alexandre Pauwels of Cambridge University independently discovered the exposed data. Two versions of the draft were found: version one named "Mythos" and version two named "Capybara," though the subtitle in both still read "Claude Mythos." Also among the leaked materials was a PDF describing an invitation-only European CEO retreat at an 18th-century English manor, with Anthropic CEO Dario Amodei in attendance. Anthropic confirmed the model's existence and attributed the exposure to "human error" in CMS configuration.

  • March 27, 2026
    Media Cascade and Market Impact Leak Event

    Fortune published its exclusive report. The same day, Bloomberg and The Information reported that Anthropic was considering an IPO as early as Q4 2026. The Information's headline read: "Anthropic Discusses Q4 IPO, Prepares Premium Claude Mythos / Capybara AI." Cybersecurity stocks began dropping sharply as investors processed the implications of Mythos's reported cybersecurity capabilities — a model that could, according to Anthropic's own leaked materials, discover vulnerabilities faster than defenders could patch them.

  • March 28, 2026
    Wider Coverage, Deeper Market Impact Leak Event

    Reporting expanded to Mashable, The Decoder, and TechZine. Market impact widened significantly — Palo Alto Networks (PANW) fell 9.8% for the week, contributing to an estimated loss of over $14.5 billion in combined market value across major cybersecurity stocks. The scale of the reaction underscored how seriously the market took Anthropic's own warning that Mythos "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders."

  • March 29, 2026
    Current Status

    Claude Mythos remains in early access testing with a small group of selected customers, with priority given to cybersecurity defense organizations. The MCP ecosystem has grown to 97 million monthly SDK downloads and over 5,800 community and enterprise servers — a staggering increase from the roughly 2 million downloads at MCP's November 2025 launch. This infrastructure will be critical to how Mythos integrates with real-world tools and workflows once it reaches broader availability.

What Comes Next

While Anthropic has been characteristically cautious about forward-looking statements, several indicators point to what the coming months may hold for Claude Mythos and the broader Claude ecosystem.

  • April 2026 (Expected)
    More Information and Possible Formal Naming Upcoming

    Anthropic has indicated that more details about Claude Mythos will be shared in April 2026. This could include formal naming confirmation (resolving the Mythos vs. Capybara question), published benchmark scores, pricing structure, and expanded access tiers. The community is watching closely for a system card or technical report that would provide the level of detail Anthropic typically releases alongside major models.

  • Coming Weeks
    Expanded Early Access Upcoming

    Anthropic has confirmed it is "slowly expanding access over the coming weeks." This phased rollout reflects both the model's high computational cost and the safety considerations around its cybersecurity capabilities. Defense-oriented organizations are expected to continue receiving priority, with broader API access following as Anthropic completes its safety evaluations and cost optimization work.

  • Date TBD
    European CEO Summit Upcoming

    Leaked materials referenced an invitation-only European CEO retreat at an 18th-century English manor where unreleased Claude capabilities would be showcased, with Dario Amodei in attendance. This event suggests Anthropic is pursuing high-touch enterprise relationships as part of the Mythos go-to-market strategy — positioning the Capybara tier for premium enterprise customers before general availability.

  • Q4 2026 (Reported)
    Potential Anthropic IPO Upcoming

    Both Bloomberg and The Information reported that Anthropic is considering an IPO as early as Q4 2026. Claude Mythos and the Capybara tier are positioned as key capability milestones that demonstrate the company's technical leadership ahead of a potential public offering. The timing would align with Anthropic demonstrating production-scale deployment of its most advanced model.

The Bigger Picture

Viewed in aggregate, the Claude timeline reveals several important patterns. First, Anthropic's release cadence has accelerated — from roughly annual generational jumps (Claude 2 to Claude 3) to multiple releases per quarter by late 2025. Second, the three-tier structure introduced with Claude 3 proved durable enough to last nearly three years before Mythos necessitated a fourth tier. Third, the mid-tier Sonnet line has consistently punched above its weight, often matching or approaching the previous generation's flagship within months of release.

The Mythos leak disrupted what would have been a carefully orchestrated reveal, but it also demonstrated something important about the current state of AI development: the capabilities described in Anthropic's own leaked drafts were significant enough to move billions of dollars in public market value within days. Whether that market reaction proves proportionate will depend on the benchmarks and real-world performance data that Anthropic is expected to publish in the coming weeks.

For now, Claude Mythos stands as the next chapter in a story that has moved from experimental AI assistant to industry-defining platform in just under three years. The timeline above will be updated as new information becomes available.