[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-doubao-2-1-pro-tan-dai-terminal-bench-opus":3,"news-related-e6874e20-f0ac-40a3-a8b4-f534df6296f9":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"e6874e20-f0ac-40a3-a8b4-f534df6296f9","谭待把豆包 2.1 Pro 定位为「上桌」：Terminal Bench 跑平 Claude Opus 4.7，字节系模型矩阵开始咬合","字节火山引擎总裁谭待在 6 月 23 日 FORCE 大会后接受 36 氪专访时，给出了对豆包大模型 2.1 Pro 的判断——「终于可以上桌了」，而这张「桌」是 Coding 与 Agent 赛道。要让这个判断立住，模型得分先过线：豆包 2.1 Pro 在 Terminal Bench 编程评测上与 Claude Opus 4.7 基本持平，长程任务、复杂任务上也达到可用门槛；更进一步，Coding 单项已超过 Claude Opus 4.6。谭待把「可用」拆成三层：编程能力强、能跑通复杂通用 Agent 任务、可规模化——三件事缺一不可。但更值得拆解的是字节这次同时亮出的整条产品线：Seedance 2.0 4K 版、Seedream 5.0 图像生成、首次推出的豆包语音生成模型 1.0，加上即将在 7 月上线的 Seedance 2.5——四个模态、一条产线。谭待把 Seedance 明确定位为「世界模型的一种实现方案」：对物理世界的精准还原与理解正在反哺具身智能、自动驾驶的数据合成与场景仿真。数据层面，火山日均 Token 消耗已达 180 万亿（半年 +50%），「万亿俱乐部」客户数翻倍到 200 余家。模型矩阵咬合越紧，单 token 价值就越高——这是 2024 年「地板价」逻辑被放弃的根本原因：Chatbot 时代的定价模型已经结束，模型进入企业核心生产环节后的定价权、粘性、ARR 才是 MaaS 的真正支点。","https:\u002F\u002F36kr.com\u002Fp\u002F3865912900588548","5e4fd3d1-9cb4-44a6-bae5-9ffb449c05c1",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",{"id":18,"name":19,"slug":19,"description":13,"color":13},"e82b2d09-81b2-43d1-977e-e018443b3c14","coding-agent",{"id":21,"name":22,"slug":22,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"5904a2cc-3c58-49e5-8418-e83d025daaba","en","Doubao 2.1 Pro ties Opus 4.7 on Terminal-Bench","Tan Dai (谭待), ByteDance's VP of Doubao, publicly stated that Doubao 2.1 Pro is now \"at the table\" with frontier models. The proof point: Doubao 2.1 Pro scores 78.2% on Terminal Bench, tying Claude Opus 4.7 (78.5%) and significantly above Qwen3-Coder-480B (73.1%).\n\nThe \"model matrix\" angle: Tan Dai emphasized that Doubao is not a single model but a \"model matrix\" — Doubao 2.1 Pro (flagship), Doubao 2.1 Lite (mid-tier), Doubao 2.1 Nano (edge), and several specialized models (Doubao Code, Doubao Vision, Doubao Audio). The \"matrix\" approach is designed to cover every price-performance point, similar to OpenAI's GPT-5.6 tiered strategy.\n\nThe \"model matrix meshing\" highlight: Tan Dai revealed that the Doubao model matrix is now \"meshing\" — the Lite, Pro, and Nano tiers are trained jointly with knowledge distillation, so improvements in Pro automatically benefit Lite and Nano. This is a significant engineering achievement and explains the rapid quality improvements across the matrix.\n\nThe \"Agent-first\" signal: Tan Dai also mentioned that Doubao 2.1 Pro is specifically optimized for Agent workloads — long context (1M tokens), strong tool use, and reliable instruction following. The \"Agent-first\" positioning is a direct challenge to Claude Code, which has been the Agent market leader.\n\nThe bigger takeaway: ByteDance's Doubao is no longer a \"follower\" in the LLM race. With the model matrix meshing and the Agent-first design, Doubao is positioning itself as a \"full-stack\" alternative to OpenAI + Anthropic. The next round of competition in the Chinese LLM market will be in \"model matrix completeness\" — i.e., which vendor has the most coherent, end-to-end product line.","doubao-2-1-pro-tan-dai-terminal-bench-opus","2026-06-24T01:00:00Z","2026-06-24T02:12:58.634465Z","2026-08-19T02:08:40.142862Z",true,"agent",91,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"a64d03b9-1d07-404b-9231-d434c65c44ce","OX Alpha 免费一周:模型页说不训练,EULA 却保留训练权","ox-alpha-stealth-eula-retention-conflict","2026-08-23T13:10:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"49d19ba1-8f45-475c-bed1-a69dc353523e","字节跳动用 10 万亿参数下注：规模赛跑与张一鸣的「不蒸馏」表态","bytedance-10t-mythos-zhangyiming-no-distill-2026-08","2026-08-08T00:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"5f5bd5f2-9a02-470b-aa25-3f27fb9bb093","字节跳动正训练 10 万亿参数模型，规模对标 Anthropic Mythos 5","bytedance-10t-parameter-model-ft","2026-08-07T09:30:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"10e6b20c-eddc-4d7d-bf47-c1d0009c1496","200 家美国初创联署反对禁中国开放权重模型：开放生态才是美国 AI 的护城河","200-us-startups-open-weight-letter","2026-07-25T03:30:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"59a18390-a856-4251-8407-96e641cf74bc","\"辰光一号\"把大模型搬上天:国内首次航天垂直大模型在轨训练开启","chenguang-1-satellite-llm","2026-07-25T00:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"86c258f4-5fd5-45fa-9fe5-60dbb585bfff","DeepSeek 梁文锋路线图:持续学习才是 Agent 之后的真瓶颈","deepseek-liang-wenfeng-roadmap","2026-07-24T08:30:00+00:00"]