{"id":"2095311586256973871","url":"https://x.com/soyaakinohara5/status/2095311586256973871","text":"今こいつの日本語強化+無検閲化したやつを頑張ってグラボ燃やしながら作ってるから待っててね","author":{"name":"秋ノ原窓也@第五拠点","username":"soyaakinohara5","avatarUrl":"https://pbs.twimg.com/profile_images/2079105510742298624/eAbf_2dg_200x200.jpg"},"createdAt":"Thu Sep 03 00:43:05 +0000 2026","engagement":{"replies":0,"retweets":2,"likes":14,"views":559},"quoteTweet":{"id":"2095039252228678071","url":"https://x.com/ModelScope2022/status/2095039252228678071","text":"Agent capabilities in just 4B parameters, with native 1M context. Don’t miss Spark-X2.5-4B. Apache 2.0. \n🤖 https://modelscope.ai/collections/XHToken/Spark-X25 \n\n🏆 Beats Qwen3.5-9B on SWE-bench Pro (44.4 vs. 33.8) and SWE-bench Multilingual (53.3 vs. 43.3).\n🛠️ Leads the comparison on τ³-bench, MCP-Atlas, MCP-Mark, Workspace Bench, VitaBench 2.0, and BrowseComp.\n💻 Fix repositories, write code, call tools, and work inside Codex, Claude Code, OpenClaw, and Hermes.\n🧮 Scores 90.7 on AIME 2026 and 81.2 on HMMT February 2026.\n🌍 Native 1M-token context, 200+ languages, and support for SGLang, vLLM, llama.cpp, MLX, Ollama, and LM Studio.","author":{"name":"ModelScope","username":"ModelScope2022","avatarUrl":"https://pbs.twimg.com/profile_images/1954838522696982528/_vUSgyLj_200x200.jpg"},"createdAt":"Wed Sep 02 06:40:56 +0000 2026","media":{"photos":[{"url":"https://pbs.twimg.com/media/HRMTC3nWwAMZz4l.jpg?name=orig","width":1107,"height":616}],"videos":[]}}}