Mystery model Agnes 3.0 Flash tops efficiency index, a 33B dense rival to Qwen
Informal-Trouble2183 · reddit · 2026-09-12
A mysterious new model, Agnes 3.0 Flash from Sapiens-AI / Agnes-AI, tops a community-built agentic coding efficiency index.
- 33B dense architecture, in the same league as Qwen3.8 27B, allegedly beating it on several benchmarks
- Weights are live on Hugging Face, but they correspond to the preview version
- The only benchmarks for the non-preview (API) version come from Artificial Analysis; final judgment awaits the full weights
The author also links to their agentic coding index at ai-leaderboard.dev.
Related event: Mysterious Open-Source 33B Model Agnes-3.0-Flash Gains Attention(3 posts)→
More from Models
- Hidden TTS feature discovered in DeepSeek web, unlocked via console snippet — teortaxesTex · 2026-09-13
- User: GLM-5.3 coding subscription beats Grok, but pair it with Opencode — thatroblennon · 2026-09-13
- Dev: Agent incidents come from labs rushing and botching RL, not from capable agents — snwy_me · 2026-09-13
- Models aren't turning evil—they're just doing what bad RL trained them to do — snwy_me · 2026-09-13
- Dev Claims GPT-6 'Astra' One-Shotted a 3D Racing Game After Watching Two Films — cedric_chee · 2026-09-13
- Tencent's AuK Space Trends on Hugging Face — tencent · 2026-09-13