Claude Mythos Nears Release, Eyes on Performance

On May 24, the identifier 'claude-mythos-1-preview' began surfacing across developer communities and X. The appearances offer the strongest signal yet that the general release of Claude Mythos — Anthropic's most powerful model, whose existence was confirmed in March — is approaching.

TestingCatalog, an outlet that tracks AI development, reported that the 'claude-mythos-1-preview' identifier is being prepared for integration with Claude Code and Claude Security. Concurrently, several users reported that the model name briefly appeared within the Claude application before disappearing hours later.

The focus has now shifted from whether the model exists to its impending availability. After spending nearly two months restricted to a small group of partners, the market is watching to see when general users will gain access and how the model's performance will measure up.

Brief Appearance of 'claude-mythos-1-preview'

The signals from May 24 center on three key developments. First, the model identifier itself. TestingCatalog noted that 'claude-mythos-1-preview' and associated UI strings appeared during an update to Claude Code and the security dashboard. That report was quickly widely shared, being retweeted more than 150 times within 24 hours.

Second, the brief exposure. Users reported that the model name was temporarily visible in the Claude backend before being removed. This mirrors a common pre-launch pattern in the industry, where models are briefly connected to production systems before an official release.

Third, a concurrent release signal. Accounts tracking the exposure noted that a second identifier, 'claude-opus-4.8', was spotted alongside the Mythos preview. This suggests that Anthropic is preparing to release both Mythos as a distinct top-tier model and a new iteration of its existing Opus line on a similar timeline.

A Four-Month Buildout and the Glasswing Detour

The preparations follow months of development. The name Mythos first became public in late March when a configuration error in Anthropic's content management system exposed approximately 3,000 internal documents. A draft blog post within that leak revealed the existence of the company's "most powerful model," leading to an exclusive report by Fortune. Anthropic subsequently confirmed it was developing a general-purpose model with "meaningful advances in reasoning, coding and cybersecurity."

Rather than moving directly to a general launch, Anthropic initially took a restricted approach. On April 7, the company provided 'Claude Mythos Preview' exclusively to a select group of partners through Glasswing, a defensive cybersecurity initiative. The strategy prioritized giving cyber defenders first access to a model deemed too powerful for immediate public release.

An update published on May 22 demonstrated the scale of that restricted phase. Roughly 50 partners tested the preview, uncovering more than 10,000 high and critical vulnerabilities and leading to 1,596 formal disclosures across 281 open-source projects. Crucially, Anthropic stated in the update that it "looks forward to making Mythos-class models available through a general release" once adequate safeguards are established. The statement marked a departure from the company's earlier stance that Mythos would remain restricted, signaling an accelerating release schedule.

Cybersecurity Remains the Primary Constraint

Despite signs of an approaching launch, cybersecurity concerns continue to dictate the release timeline. In the draft document leaked in March, Anthropic described Mythos as being "currently far ahead of any other AI model in cyber capabilities."

The same document cautioned that the model "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders." Concerns that the model could facilitate large-scale cyberattacks if misused have driven Anthropic's strategy of deploying it to defensive organizations first.

Such caution aligns with previous capabilities demonstrated by Anthropic's models. Last year, Opus 4.6 proved capable of identifying unknown vulnerabilities in production codebases, a feature the company acknowledged as having dual-use potential for both offense and defense. Mythos is expected to possess significantly more advanced capabilities in this area.

Delay Speculation Points to Cost and Distillation

Beyond safety concerns, industry speculation on May 24 centered on operational costs as a factor in the delayed general release. One widely circulated developer theory suggested that Anthropic is attempting to distill Mythos into a smaller, more efficient model that can run at a massive scale with lower operating costs.

While unverified, the theory aligns with internal documents. The March draft noted that Mythos was expensive to operate and not yet ready for general availability. Refining a lighter version that maintains core performance while reducing serving costs would address those financial constraints.

The operational burden is substantial. Industry estimates suggest Anthropic spends approximately $1.25 billion monthly on 300MW of power strictly for GPU operations. Additional reports from Japan indicated that the company had previously restricted Mythos to just six firms due to its power, further reinforcing the perception that general release involves hurdles beyond standard safety reviews.

Anticipating Performance Beyond the Opus Tier

Once those hurdles are cleared, the market's primary focus will be on the model's actual capabilities. In the March draft, Anthropic referred to the model under the tier name "Capybara," describing it as "a new tier larger and more intelligent than Opus." The document claimed the model achieved dramatically higher scores than Opus 4.6 on software coding, academic reasoning, and cybersecurity benchmarks.

The dual internal names, 'Capybara' and 'Mythos,' suggest a significant shift in the company's product structure. Positioning a new grade above the existing Opus-Sonnet-Haiku lineup indicates a broader reshaping of Anthropic's offerings rather than a standard version update. The simultaneous appearance of the 'claude-opus-4.8' identifier supports this view, with rumors suggesting the new Opus iteration could launch in the coming weeks, maintaining the cadence set by Opus 4.7 in April.

Mythos is not the only frontier model generating release rumors. Speculation surrounding Google's next Gemini iteration has also intensified, highlighting a rapidly moving competitive landscape (related article). While Anthropic has not yet confirmed a general release date for Mythos, market attention is already pivoting toward how far the new model will exceed the capabilities of the current Opus tier.