Anthropic、開発者会議前に新モデル「Jupiter-v1-p」のテストを開始

Anthropic appears to have started a fresh round of red teaming on a new internal build referred to as Claude Jupiter V1. The codename follows a pattern the company has used before: pre-release safety probes are tagged with planet names rather than the eventual product label, keeping the actual designation hidden until launch. Jupiter V1 is purely an internal handle and is not expected to surface in any public API string or product UI.

The timing is what makes this worth flagging. Anthropic is set to host its Code with Claude developer conference in San Francisco on May 6th, with London and Tokyo dates following later in the spring. Last year, the company ran an analogous exercise under the codename Neptune just weeks before unveiling the Claude 4 family, wrapping up in mid-May 2025, a sequence that served as a soft signal for the launch that followed. A repeat of that pattern would suggest Jupiter V1 is being hardened ahead of an announcement timed to the SF event.

The current lineup has Opus 4.7 as the flagship, while Sonnet 4.7 and Haiku 4.7 are conspicuously absent, leaving room for either a refresh of the mid- and small-tier models or a step up to a new generation built on the Mythos foundation that has surfaced in earlier reporting. The planet metaphor invites its own speculation: Jupiter is the largest body in the solar system, and the symbolism has not gone unnoticed.

If released on schedule, the model would surface across the Anthropic Platform, Claude Code, and Claude consumer apps, the same channels through which Opus 4.7 rolled out earlier this spring. The red team round itself is consistent with Anthropic's responsible scaling policy, which calls for jailbreak probes and constitutional classifier stress tests before any frontier-class deployment. Whether the May 6th keynote brings a full new generation or a measured 4.7 rollout for Sonnet and Haiku, or something else, the precedent set by Neptune suggests an announcement is close.