Claude Sonnet 5 Fennec Just Leaked — Here's What We Know

A recent Vertex AI error log has surfaced, indicating the existence of a previously unknown model, claude-sonnet-5, internally codenamed "Fennec." This log, timestamped February 3, 2026, aligns with Anthropic's established model versioning conventions. The leak suggests that Sonnet 5 is likely trained on Google TPUs, consistent with Anthropic's partnership with Google Cloud. Benchmark performance figures mentioned in the leak, if accurate, would indicate Sonnet 5 achieving current Opus-level capabilities at a likely lower price point. The leak also points to a "Dev Team Mode" utilizing parallel sub-agents, an extension of existing agentic workflows in Claude Code. Speculation about Sonnet 5 being "50% cheaper than Opus" or "one generation ahead of Snow Bunny" lacks concrete evidence. The release timing is also questioned, as Anthropic recently launched Opus 4.5 and typically does not surprise market with flagship model releases. This leak occurs shortly after Anthropic CEO Dario Amodei’s disclosure regarding "alignment faking" in Claude 4 Opus, suggesting a focus on AI deception. If the leak proves true, it implies potential significant cost efficiencies for developers and a maturation of AI teams as a development paradigm. While the leak is plausible, its verification and release timeline remain uncertain, prompting cautious optimism among AI developers.