AI Briefing
KOSign in

Agnes AI Releases Agnes-3.0-Qwen: 180B MoE Model with 1M Context and Strong Agentic Benchmarks

·2026.10.21 02:11

Key point

The 180B parameter hybrid attention MoE model features a 1,048,576-token context window and selectable reasoning effort levels.

Details

  • Agnes AI released agnes-3.0-qwen, a multimodal reasoning model with approximately 180B total parameters (512 experts, 10 active per token) and a 1,048,576-token context window.
  • The model scores highest in its comparison set on τ³-Banking and is within about one point of the best model on Terminal-Bench 2.1, despite being smaller than most flagship systems compared.
  • On IFBench, agnes-3.0-qwen outperforms GPT-5.6 Sol (Max) and Claude Opus 4.8 (Max).
  • Supports text, image, and video input with text output, tool calling, and streaming in BF16 precision.
  • Released in October 2026 under the Qwen Community License 1.0, supporting English and Chinese with a maximum output of 65,536 tokens.
  • Offers selectable reasoning effort levels: xhigh (default), medium, and low.
open-sourcemultimodalmoeagentic-ailong-contextagnes-aiagnes-3.0-qwen
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

Redditllama.cpp Adds MTP Support for Mellum2.1 MoE ModelRedditinclusionAI Releases Ming-3.0-audio-tiny, a 9.3B Parameter Native Audio-Language ModelHuggingFace BlogReflection releases Beam: a 501B MoE open model with 1M context and specialized agentic capabilitiesGeekNewsArtCraft releases open-source IDE for AI image and video creation with 62-model catalogDeepSeekDeepSeek Releases V4.1-Flash-1017 and Opens Search & Fetch APIsPyTorchKRStrata Engine Released to Run Qwen3.8-Flash-Next 125B on 12GB VRAM