Nicely, I’m unsure who noticed this coming,
However Qwen3 is right here.
Not a paid member? Read here for Free!
I believe most of us have been anticipating Deepseek to drop one thing new, however plot twist.
It’s one other Chinese language open-source AI mannequin that’s surprisingly good.
The flagship mannequin is the Qwen3–235B-A22B. Qwen3 is the household of fashions, with the 235B being the entire parameters.
It’s a combination of consultants mannequin the place solely about 22B parameters are activated for any given question.
The large headline?
This factor is aggressive with top-tier fashions like DeepSeek R1, o3 Mini, Grok 3, and Gemini 2.5 Professional.
With Qwen3, you may toggle between prolonged pondering mode (reasoning) and common mode.
benchmarks, Qwen3 beats o3 Mini and comes very near Gemini 2.5 Professional on ArenaHard. On AIME 24 and 25, it sits between Gemini 2.5 Professional and o3 Mini. For LiveCodeBench…
