Upstage 的 Solar Pro 4 是首个登上多个 Arena 榜单的韩国模型，适合做生产级智能体，价格便宜，值得试试。
Upstage 发布 Solar Pro 4，成为首个登上 Agent Arena、Code Arena: WebDev 和 Text Arena 榜单的韩国实验室模型。Solar Pro 4 在 Agent Arena 排名第 44，与 MiniMax M2.7 和 Nemotron 3 Ultra 并列，整体净改进得分为 -12.10%。该模型在 GDPval-AA 得分为 39，τ³-Banking 为 23，Terminal-Bench 2.1 为 57，定价为每百万 tokens 0.30 美元（输入）和 1.20 美元（输出），缓存价格 0.06 美元，优惠 90% 至 9 月 10 日。
Solar Pro 4 by @upstageai is the first Korean Lab to hit the Arena leaderboards for Agent Arena, Cod...
Solar Pro 4 by @upstageai is the first Korean Lab to hit the Arena leaderboards for Agent Arena, Code Arena: WebDev, and Text Arena! Solar Pro 4 debuts in the Agent Arena at #44 on par with models like: MiniMax M2.7 and Nemotron 3 Ultra. Agent Arena uses causal tracing methodology to measure a model's net improvement across real-world, long-horizon agentic workflows. It indicates how much it improves outcomes relative to the average model. Solar Pro 4 has an overall-net improvement score of -12.10%. By signal it landed: - Confirmed Success: -7.3% - Praise vs. Complaint: -23.4% - Steerability: -13.6% - Bash Recovery: -16.5% - Tool Hallucination: +0.3% Congrats to @upstageai for their contribution to the AI ecosystem! Upstage @upstageai Solar Pro 4 is live. Built for one job: production agents that finish the work. ▸ GDPval-AA (real work deliverables): 39 ▸ τ³-Banking (policy retrieval + tool use): 23 ▸ Terminal-Bench 2.1: 57 🪙 $0.30 / $1.20 per 1M tokens · $0.06 cached 90% off on Upstage Console and OpenRouter through Sep 10. How it’s trained to finish jobs → go.upstage.ai/4cy3VPz g Your browser does not support the video tag. 🔗 View on Twitter 🔗 View Quoted Tweet 💬 3 🔄 0 ❤️ 9 👀 2360 📊 3 ⚡