Developers have uncovered references to a new Anthropic model codenamed “Fennec” in Google Vertex AI logs, prompting speculation that Claude Sonnet 5 could launch as early as February 3, 2026.
The discovery occurred over the weekend when the model identifier claude-sonnet-5@20260203 appeared in a misconfigured Vertex AI deployment log. The date string in the identifier matches February 3, 2026. This follows Anthropic’s versioning pattern, as seen with Claude Opus 4.5 identified as claude-opus-4-5@20251101, which corresponds to its November 2025 checkpoint.
Anthropic has issued no official announcement. No press release, product page, or documentation mentions Sonnet 5, its architecture, pricing, performance, or availability. The original source is a Twitter screenshot without verifiable access to the logs, according to tech site NerdSchalk: “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.”
Online claims, which remain unverified, state Sonnet 5 could cost 50 percent less than Claude Opus 4.5, support a one-million-token context window, and include a “Dev Team” mode for spawning specialized sub-agents in parallel development tasks. Rumors also suggest the model exceeds 80.9 percent on SWE-Bench, matching or surpassing Opus 4.5’s score from its November 2025 launch.
The speculated release aligns with Super Bowl LX on February 8, 2026, at Levi’s Stadium in Santa Clara, California. AI companies have timed product launches to coincide with such events for greater exposure.
TestingCatalog reported early hands-on tests of a non-thinking Sonnet 5 variant, which showed competitive math performance against frontier models and stronger coding output than Claude Opus 4.5 in some workflows. The build featured a 128k context window, though final specifications are unknown.
Industry observers advise caution. Anthropic’s October 2025 agreement with Google provides access to up to one million TPUs, a verified detail. Performance and pricing claims, however, stem from community speculation.
The landscape includes recent leaks of Google’s Gemini 3.5, codenamed “Snow Bunny,” and reports of OpenAI preparing GPT-5.2 upgrades. One developer stated, “Every major AI lab is in a sprint. Leaks—real or fabricated—create market pressure.”
