AAAbhimanyu ARYAN@theabhimanyu2:54 AM · Aug 22, 2026

@arena @petergostev Thanks to OpenAI's Prism for writing LaTeX for me, Antigravity for running benchmark reports and monitoring 24/7, and Copilot Harness for unlimited access to GPT-4o/5 mini models (for students).

TAThe AI Therapist@TheAIShrink7:15 PM · Aug 21, 2026

@arena @petergostev Qwen3.8-27B beating 100x larger models? that's the open-weight model finally eating the closed-weight lunch. efficiency > parameter bloat

HNHenry Nguyen@henryhndev11:20 PM · Aug 21, 2026

@arena @petergostev The "up to 100x its size" part is wild—can’t wait to see how it holds up on agentic stuff, that’s where small models usually crumble for me.

ARArena.ai@arena7:08 PM · Aug 21, 2026

370TECH

Check out first impressions of Qwen3.8-27B with @petergostev on our YouTube. We tested it against models up to 100x its size (DeepSeek v4, Qwen 3.8 Max, Kimi K3, GLM 5.3, Grok 4.6, GPT-5.6, and Fable) on identical one-shot generation + agentic tasks. Scores coming soon.

Qwen 3.8 27B | First impressionsFrom youtu.be