A small new AI model called VibeThinker-3B, made by Sina (the Chinese company that owns Weibo), performs as well as much larger models on hard math and coding tests despite having only three billion parameters. On some tests, it keeps up with models that are 200 to 333 times bigger. On a coding test called LiveCodeBench, it beats every other model under 20 billion parameters. On math (AIME26), it is on par with DeepSeek V3.2 and Kimi K2.5. To avoid data contamination, the team entered it into live LeetCode coding contests that ran from late April to late May 2026, where VibeThinker-3B solved 123 of 128 problems on its first try, putting it ahead of GPT-5.2, Qwen3-Max, Kimi K2.5, and Claude Opus 4.6. Only three models did better: GPT-5.3-Codex, Gemini 3.1 Pro, and Gemini 3 Flash. However, on the GPQA-Diamond test of general knowledge questions, the model falls well behind the bigger ones.