Ready to meet the biggest, brainiest guy in the Qwen3 family?

Upgrades to free tier in ChatGPT. Nice that Android has been prioritized.

350M model for English/Japanese translation with great performance.

A new stealth model 👀 Carrot 🥕

Branch conversations available in ChatGPT. Branching was already possible but this is more convenient.

Kimi K2-0905 update 👀

New stealth models, Sonoma Alpha 👀

Qwen3-Max-Preview available on anycoder. 💻 A shame it's not an open source model but shout-out to the Qwen team for releasing so many great open source models in the past.

New Qwen-3-Next-80B-A3B model incoming "Built on this architecture, we trained and open-sourced Qwen3-Next-80B-A3B — 80B total parameters, only 3B active — achieving extreme sparsity and efficiency. Despite its ultra-efficiency, it outperforms Qwen3-32B on downstream tasks —

ERNIE-4.5-21B-A3B-Thinking a text MoE post-trained model, with 21B total parameters and 3B activated parameters for each token. Significantly improved performance on reasoning tasks, including logical reasoning, mathematics, science, coding, text generation, and academic

I like the new wave of models with a medium to high total parameter count and a low active parameter counts like Qwen3 and ERNIE-4.5. These models are perfect to run in systems like AMD Ryzen Strix Halo with 96GB or 128GB of RAM.

MobileLLM-R1 is a 950M parameters reasoning model, good in benchmarks against similarly sized models. Small language models keep improving!

Qwen3-next-80B-A3B 👀

Goodbye to 2025 👋 Happy New Year 2026! Small LLMs highlights in 2025: - Qwen3-4B - LFM2–8B-A1B / LFM2-2.6B-Exp - granite-4.0-h-tiny (7B A1B) - gemma-3n-E4B / gemma-3-4b-it - SmolLM3-3B - Ministral-3-3B They run at decent speeds on the CPU! Looking forward to 2026's smol LLMs!

Gotta try the Trinity models 👀

Great improvements on small models by Liquid AI 👀

For us at Liquid, the goal was always to fundamentally bring efficiency to the point that you can take it for granted: models that run light, fast, on any processor. The art is can we obtain this efficiency for small models with the quality and reliability that can power real

Here was GPT 5.3 on max settings - attempt at cloning a pokemon game!