Developers uncovered references to Anthropic’s new model codenamed Fennec in Google Vertex AI logs, indicating Claude Sonnet 5 may launch on February 3, 2026. The finding emerged from a misconfigured deployment log showing the identifier:claude-sonnet-5@20260203.
The discovery occurred over the weekend. The model identifierclaude-sonnet-5@20260203appeared in the log, with the date string 20260203 corresponding exactly to February 3, 2026. This alignment prompted discussions about an imminent release from Anthropic.
The identifier adheres to Anthropic’s model versioning convention. For comparison, Claude Opus 4.5 uses the designation claude-opus-4-5@20251101, which reflects its checkpoint date in November 2025. This pattern lends credibility to the Sonnet 5 reference within Anthropic’s naming structure.
Anthropic has issued no official statement on the matter. Tech site NerdSchalk reported, “As of now, Anthropic hasn’t published a press release, product page, or documentation announcing Sonnet 5, its architecture, pricing, performance, or availability for developers or users.” No details on these aspects have surfaced through official channels.
The initial leak originated from a Twitter screenshot. This image lacks proof of direct access to the underlying Vertex AI logs, raising questions about its authenticity.
Online discussions include unverified details about Sonnet 5 capabilities. Claims state the model could cost 50 percent less than Claude Opus 4.5. Other speculations mention a one-million-token context window and a “Dev Team” mode. This mode would enable the AI to generate specialized sub-agents for handling parallel development tasks.
Rumors extend to benchmark performance. Sources suggest Sonnet 5 achieves over 80.9 percent on SWE-Bench. This score would equal or surpass the 80.9 percent recorded by Opus 4.5 upon its launch in November 2025.
The speculated timing coincides with Super Bowl LX, set for February 8, 2026, at Levi’s Stadium in Santa Clara, California. AI companies have adopted strategies of announcing products during such high-profile events to gain broad visibility.
TestingCatalog conducted early hands-on evaluation of a non-thinking Sonnet 5 variant. Results showed math performance competitive with leading frontier models. Coding output proved stronger than Claude Opus 4.5 in certain workflows. The tested build included a 128k context window, though final specifications remain undisclosed.
Observers advise restraint in interpreting the claims. Anthropic’s October 2025 agreement with Google provides confirmed access to up to one million TPUs. In contrast, specifics on Sonnet 5 performance and pricing stem from community speculation rather than verified data.
Recent developments in the sector add context. Leaked references to Google’s Gemini 3.5, codenamed “Snow Bunny,” surfaced last week. OpenAI prepares upgrades for GPT-5.2. One developer stated, “Every major AI lab is in a sprint.” The same source added, “Leaks—real or fabricated—create market pressure.”
