Leaks over the weekend point to a new Anthropic model branded as Claude Sonnet 5, with an internal date string of February 3, 2026. Whether that date reflects a public launch or an internal checkpoint remains unclear, but the timing would place it in the same week as Super Bowl LX on February 8, when AI labs have been ramping up mainstream marketing to chase consumer mindshare against ChatGPT and Google’s Gemini.
One-shot game by Claude Sonnet 5 (non thinking) h/t vvirtr
Early hands-on testing described in TestingCatalog suggests the non-thinking Sonnet 5 variant already looks competitive on math with today’s frontier-tier models, while also showing even stronger coding output than Claude Opus 4.5 in some workflows. One recurring example was structured visual generation: an ASCII world map prompt reportedly produced the most complete, detailed result the tester has seen, alongside other UI and rendering-oriented code tasks.
Newsletter landing page via Claude Sonnet 5 (non thinking) h/t vvirtr
The build observed in testing was listed with a 128k context window (context window in the release version will likely be different) and is expected to target a faster, lower-cost tier than Opus. If that pricing and latency positioning holds, Sonnet 5 would fit Anthropic’s recent pattern of shipping a broadly usable “workhorse” model for developers and everyday users, while OpenAI and Google compete for the same audience.
Outputs were provided by vvirtr
