DeepSeek发布的V4 - Pro (High)，比前代版本强，成本还低，在开放模型里排第二，可以去试试。
DeepSeek - V4 - Pro (High) 以+6%净提升成为开放模型领域Agent Arena第2名；与DeepSeek - V4 - Flash (High) 相比，其性能更高且任务成本更低；在代码和工作类别排名第2，聊天类第5，多项指标表现突出。
Exciting update: DeepSeek-V4-Pro (High) by @deepseek_ai is now #2 among open models in Agent Arena (...
Exciting update: DeepSeek-V4-Pro (High) by @deepseek_ai is now #2 among open models in Agent Arena ( #16 overall), with +6% net improvement! At $0.21 median cost/task, it reshapes the Agent Arena Pareto frontier! Compared to DeepSeek-V4-Flash (High), it has higher performance with a lower median cost per task. Among open models, it lands two spots above DeepSeek-V4-Flash (High), and by category, it’s #2 in Code and Work, and #5 in Chat. By signal, DeepSeek-V4-Pro (High) is strong in Confirmed Success (“yes that worked” from the user) with +12.7%, and in Bash Recovery (CLI error recovery) with +10.9%. - Bash Recovery: #1 open / #13 overall (+10.9%) - Confirmed Success: #2 open / #4 overall (+12.7%) - Praise vs Complaint: #4 open / #20 overall (+3.4%) - Steerability: #4 open / #25 overall (+1.7%) - No issues with Tool Hallucination (+1.1%) DeepSeek @deepseek_ai We’re launching DeepSeek-V4-Pro today! 🚀 🔷 Major Agent upgrades with strong production gains! 🔷 Flexible reasoning effort for V4-Pro & V4-Flash: low for simple tasks, high for daily Agent workflows, max for complex tasks. 🔷 Native OpenAI Responses API support, optimized for Codex with one-click setup. V4 Pro is now available on app/web. Try it via “Expert Mode”. V4 Pro is also available via API. Model names remain unchanged—please refer to the API docs for setup details. 🔗 View Quoted Tweet 💬 4 🔄 11 ❤️ 68 👀 7781 📊 11 ⚡