[{"data":1,"prerenderedAt":1096},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"model-deepseek-r1":9},{"tools":4,"reviews":5},65,7,{"tools":4,"reviews":5,"playbooks":7,"news":8},10,8,{"id":10,"title":11,"apiCompatible":12,"benchmarks":14,"body":27,"category":1062,"contextWindow":1063,"description":1064,"extension":1065,"maxOutput":1066,"meta":1067,"navigation":433,"path":1068,"pricing":1069,"published":1070,"relatedTools":1071,"releaseDate":1074,"seo":1075,"slug":1076,"stem":1077,"strengths":1078,"updated":1070,"useCases":1084,"vendor":1088,"vendorEn":1089,"weaknesses":1090,"__hash__":1095},"models\u002Fmodels\u002Fdeepseek-r1.md","DeepSeek-R1",[13],"openai",[15,18,21,24],{"name":16,"score":17},"MATH-500","97.3%",{"name":19,"score":20},"AIME 2024","79.8%",{"name":22,"score":23},"GPQA Diamond","58.2%",{"name":25,"score":26},"HumanEval","89.2%",{"type":28,"value":29,"toc":1045},"minimark",[30,34,38,41,44,48,51,135,138,141,144,269,275,304,307,314,364,367,371,590,594,607,668,671,674,712,719,722,726,817,822,833,837,915,922,925,928,954,957,971,974,1006,1009,1041],[31,32,33],"h2",{"id":33},"概述",[35,36,37],"p",{},"DeepSeek-R1 是深度求索于 2025 年 1 月与 V3 同步发布的推理模型。与 V3 的区别在于：R1 在回答前会先\"想一想\"（思维链），在数学、逻辑、科学推理上远超 V3。",[35,39,40],{},"R1 的发布是开源大模型领域的一个分水岭事件——首次让\"推理模型\"以完全开源 + 思维链可见的形式进入业界，比 OpenAI o1（思维链黑盒）更开放。",[31,42,43],{"id":43},"核心能力",[45,46,47],"h3",{"id":47},"推理能力",[35,49,50],{},"R1 在 MATH-500 上拿到 97.3%，AIME 2024（美国数学竞赛）79.8%。这些成绩接近 GPT-5（98.4% \u002F 82.3%），远超非推理模型：",[52,53,54,68],"table",{},[55,56,57],"thead",{},[58,59,60,64,66],"tr",{},[61,62,63],"th",{},"模型",[61,65,16],{},[61,67,19],{},[69,70,71,83,91,102,113,124],"tbody",{},[58,72,73,77,80],{},[74,75,76],"td",{},"GPT-5",[74,78,79],{},"98.4%",[74,81,82],{},"82.3%",[58,84,85,87,89],{},[74,86,11],{},[74,88,17],{},[74,90,20],{},[58,92,93,96,99],{},[74,94,95],{},"Claude Opus 4",[74,97,98],{},"~96%",[74,100,101],{},"~74%",[58,103,104,107,110],{},[74,105,106],{},"DeepSeek-V3",[74,108,109],{},"90.2%",[74,111,112],{},"39.2%",[58,114,115,118,121],{},[74,116,117],{},"Claude Sonnet 4",[74,119,120],{},"92%",[74,122,123],{},"49%",[58,125,126,129,132],{},[74,127,128],{},"GPT-4o",[74,130,131],{},"76.6%",[74,133,134],{},"13.4%",[35,136,137],{},"注意 V3 → R1 在 AIME 上从 39% 飙升到 79%——这就是\"推理模式\"带来的差距。",[45,139,140],{"id":140},"思维链可见",[35,142,143],{},"R1 的思维链完全开放——你可以看到模型一步步推理的过程：",[145,146,151],"pre",{"className":147,"code":148,"language":149,"meta":150,"style":150},"language-python shiki shiki-themes github-light github-dark","resp = client.chat.completions.create(\n    model=\"deepseek-reasoner\",   # R1 的别名\n    messages=[{\"role\": \"user\", \"content\": \"证明...\"}],\n)\n# 关键字段\nprint(resp.choices[0].message.reasoning_content)   # 思维链\nprint(resp.choices[0].message.content)             # 最终回答\n","python","",[152,153,154,170,190,224,230,236,255],"code",{"__ignoreMap":150},[155,156,159,163,167],"span",{"class":157,"line":158},"line",1,[155,160,162],{"class":161},"sVt8B","resp ",[155,164,166],{"class":165},"szBVR","=",[155,168,169],{"class":161}," client.chat.completions.create(\n",[155,171,173,177,179,183,186],{"class":157,"line":172},2,[155,174,176],{"class":175},"s4XuR","    model",[155,178,166],{"class":165},[155,180,182],{"class":181},"sZZnC","\"deepseek-reasoner\"",[155,184,185],{"class":161},",   ",[155,187,189],{"class":188},"sJ8bj","# R1 的别名\n",[155,191,193,196,198,201,204,207,210,213,216,218,221],{"class":157,"line":192},3,[155,194,195],{"class":175},"    messages",[155,197,166],{"class":165},[155,199,200],{"class":161},"[{",[155,202,203],{"class":181},"\"role\"",[155,205,206],{"class":161},": ",[155,208,209],{"class":181},"\"user\"",[155,211,212],{"class":161},", ",[155,214,215],{"class":181},"\"content\"",[155,217,206],{"class":161},[155,219,220],{"class":181},"\"证明...\"",[155,222,223],{"class":161},"}],\n",[155,225,227],{"class":157,"line":226},4,[155,228,229],{"class":161},")\n",[155,231,233],{"class":157,"line":232},5,[155,234,235],{"class":188},"# 关键字段\n",[155,237,239,243,246,249,252],{"class":157,"line":238},6,[155,240,242],{"class":241},"sj4cs","print",[155,244,245],{"class":161},"(resp.choices[",[155,247,248],{"class":241},"0",[155,250,251],{"class":161},"].message.reasoning_content)   ",[155,253,254],{"class":188},"# 思维链\n",[155,256,257,259,261,263,266],{"class":157,"line":5},[155,258,242],{"class":241},[155,260,245],{"class":161},[155,262,248],{"class":241},[155,264,265],{"class":161},"].message.content)             ",[155,267,268],{"class":188},"# 最终回答\n",[35,270,271,274],{},[152,272,273],{},"reasoning_content"," 字段在 OpenAI 兼容接口上是 DeepSeek 的扩展。这对以下场景特别有价值：",[276,277,278,286,292,298],"ul",{},[279,280,281,285],"li",{},[282,283,284],"strong",{},"教育场景"," — 学生可以看到解题思路",[279,287,288,291],{},[282,289,290],{},"调试场景"," — 开发者可以理解模型为什么这样回答",[279,293,294,297],{},[282,295,296],{},"信任建立"," — 可验证的推理过程",[279,299,300,303],{},[282,301,302],{},"数据蒸馏"," — 用 R1 的思维链训练小模型（Meta 这么干过）",[45,305,306],{"id":306},"开源",[35,308,309,310,313],{},"R1 模型权重完全开源（MIT 协议）。可以在自己的 GPU 上部署，不依赖 API。同时 DeepSeek 也放出了 ",[282,311,312],{},"R1-Distill"," 系列——把 R1 的能力蒸馏到 Llama \u002F Qwen 的小模型上：",[52,315,316,329],{},[55,317,318],{},[58,319,320,323,326],{},[61,321,322],{},"蒸馏版本",[61,324,325],{},"显存需求",[61,327,328],{},"性能保留",[69,330,331,342,353],{},[58,332,333,336,339],{},[74,334,335],{},"R1-Distill-Qwen-1.5B",[74,337,338],{},"4GB",[74,340,341],{},"数学接近 V3",[58,343,344,347,350],{},[74,345,346],{},"R1-Distill-Qwen-7B",[74,348,349],{},"16GB",[74,351,352],{},"推理接近 GPT-4o",[58,354,355,358,361],{},[74,356,357],{},"R1-Distill-Llama-70B",[74,359,360],{},"140GB",[74,362,363],{},"接近原版 R1",[35,365,366],{},"1.5B 在笔记本上就能跑推理模型——这是 R1 之前完全不可能的。",[31,368,370],{"id":369},"api-调用示例","API 调用示例",[145,372,374],{"className":147,"code":373,"language":149,"meta":150,"style":150},"from openai import OpenAI\nclient = OpenAI(\n    api_key=\"sk-...\",\n    base_url=\"https:\u002F\u002Fapi.deepseek.com\u002Fv1\",\n)\n\nresp = client.chat.completions.create(\n    model=\"deepseek-reasoner\",\n    messages=[\n        {\"role\": \"user\", \"content\": \"证明素数无穷\"}\n    ],\n    # 注意：R1 不支持 temperature \u002F top_p \u002F presence_penalty 等参数\n    # 传了会被忽略\n    max_tokens=8000,\n)\n\nmsg = resp.choices[0].message\nprint(\"【思考过程】\")\nprint(msg.reasoning_content)\nprint(\"\\n【最终答案】\")\nprint(msg.content)\n",[152,375,376,390,400,413,425,429,435,443,453,463,486,492,498,504,517,522,527,543,556,564,582],{"__ignoreMap":150},[155,377,378,381,384,387],{"class":157,"line":158},[155,379,380],{"class":165},"from",[155,382,383],{"class":161}," openai ",[155,385,386],{"class":165},"import",[155,388,389],{"class":161}," OpenAI\n",[155,391,392,395,397],{"class":157,"line":172},[155,393,394],{"class":161},"client ",[155,396,166],{"class":165},[155,398,399],{"class":161}," OpenAI(\n",[155,401,402,405,407,410],{"class":157,"line":192},[155,403,404],{"class":175},"    api_key",[155,406,166],{"class":165},[155,408,409],{"class":181},"\"sk-...\"",[155,411,412],{"class":161},",\n",[155,414,415,418,420,423],{"class":157,"line":226},[155,416,417],{"class":175},"    base_url",[155,419,166],{"class":165},[155,421,422],{"class":181},"\"https:\u002F\u002Fapi.deepseek.com\u002Fv1\"",[155,424,412],{"class":161},[155,426,427],{"class":157,"line":232},[155,428,229],{"class":161},[155,430,431],{"class":157,"line":238},[155,432,434],{"emptyLinePlaceholder":433},true,"\n",[155,436,437,439,441],{"class":157,"line":5},[155,438,162],{"class":161},[155,440,166],{"class":165},[155,442,169],{"class":161},[155,444,445,447,449,451],{"class":157,"line":8},[155,446,176],{"class":175},[155,448,166],{"class":165},[155,450,182],{"class":181},[155,452,412],{"class":161},[155,454,456,458,460],{"class":157,"line":455},9,[155,457,195],{"class":175},[155,459,166],{"class":165},[155,461,462],{"class":161},"[\n",[155,464,465,468,470,472,474,476,478,480,483],{"class":157,"line":7},[155,466,467],{"class":161},"        {",[155,469,203],{"class":181},[155,471,206],{"class":161},[155,473,209],{"class":181},[155,475,212],{"class":161},[155,477,215],{"class":181},[155,479,206],{"class":161},[155,481,482],{"class":181},"\"证明素数无穷\"",[155,484,485],{"class":161},"}\n",[155,487,489],{"class":157,"line":488},11,[155,490,491],{"class":161},"    ],\n",[155,493,495],{"class":157,"line":494},12,[155,496,497],{"class":188},"    # 注意：R1 不支持 temperature \u002F top_p \u002F presence_penalty 等参数\n",[155,499,501],{"class":157,"line":500},13,[155,502,503],{"class":188},"    # 传了会被忽略\n",[155,505,507,510,512,515],{"class":157,"line":506},14,[155,508,509],{"class":175},"    max_tokens",[155,511,166],{"class":165},[155,513,514],{"class":241},"8000",[155,516,412],{"class":161},[155,518,520],{"class":157,"line":519},15,[155,521,229],{"class":161},[155,523,525],{"class":157,"line":524},16,[155,526,434],{"emptyLinePlaceholder":433},[155,528,530,533,535,538,540],{"class":157,"line":529},17,[155,531,532],{"class":161},"msg ",[155,534,166],{"class":165},[155,536,537],{"class":161}," resp.choices[",[155,539,248],{"class":241},[155,541,542],{"class":161},"].message\n",[155,544,546,548,551,554],{"class":157,"line":545},18,[155,547,242],{"class":241},[155,549,550],{"class":161},"(",[155,552,553],{"class":181},"\"【思考过程】\"",[155,555,229],{"class":161},[155,557,559,561],{"class":157,"line":558},19,[155,560,242],{"class":241},[155,562,563],{"class":161},"(msg.reasoning_content)\n",[155,565,567,569,571,574,577,580],{"class":157,"line":566},20,[155,568,242],{"class":241},[155,570,550],{"class":161},[155,572,573],{"class":181},"\"",[155,575,576],{"class":241},"\\n",[155,578,579],{"class":181},"【最终答案】\"",[155,581,229],{"class":161},[155,583,585,587],{"class":157,"line":584},21,[155,586,242],{"class":241},[155,588,589],{"class":161},"(msg.content)\n",[45,591,593],{"id":592},"multi-turn-注意事项","Multi-turn 注意事项",[35,595,596,599,600,602,603,606],{},[282,597,598],{},"重要","：multi-turn 对话时不要把 ",[152,601,273],{}," 加回 messages 历史，只保留 ",[152,604,605],{},"content","：",[145,608,610],{"className":147,"code":609,"language":149,"meta":150,"style":150},"# ❌ 错误：把思维链塞回历史，下一轮上下文翻倍\nhistory.append({\"role\": \"assistant\", \"content\": msg.reasoning_content + msg.content})\n\n# ✅ 正确：只保留最终答案\nhistory.append({\"role\": \"assistant\", \"content\": msg.content})\n",[152,611,612,617,642,646,651],{"__ignoreMap":150},[155,613,614],{"class":157,"line":158},[155,615,616],{"class":188},"# ❌ 错误：把思维链塞回历史，下一轮上下文翻倍\n",[155,618,619,622,624,626,629,631,633,636,639],{"class":157,"line":172},[155,620,621],{"class":161},"history.append({",[155,623,203],{"class":181},[155,625,206],{"class":161},[155,627,628],{"class":181},"\"assistant\"",[155,630,212],{"class":161},[155,632,215],{"class":181},[155,634,635],{"class":161},": msg.reasoning_content ",[155,637,638],{"class":165},"+",[155,640,641],{"class":161}," msg.content})\n",[155,643,644],{"class":157,"line":192},[155,645,434],{"emptyLinePlaceholder":433},[155,647,648],{"class":157,"line":226},[155,649,650],{"class":188},"# ✅ 正确：只保留最终答案\n",[155,652,653,655,657,659,661,663,665],{"class":157,"line":232},[155,654,621],{"class":161},[155,656,203],{"class":181},[155,658,206],{"class":161},[155,660,628],{"class":181},[155,662,212],{"class":161},[155,664,215],{"class":181},[155,666,667],{"class":161},": msg.content})\n",[35,669,670],{},"这点新手最容易踩——把思维链当成\"模型记忆\"塞回去，结果上下文成本飞涨且模型困惑。",[31,672,673],{"id":673},"定价",[52,675,676,686],{},[55,677,678],{},[58,679,680,683],{},[61,681,682],{},"项目",[61,684,685],{},"价格",[69,687,688,696,704],{},[58,689,690,693],{},[74,691,692],{},"Input",[74,694,695],{},"¥1 \u002F 百万 token",[58,697,698,701],{},[74,699,700],{},"Input（缓存命中）",[74,702,703],{},"¥0.1 \u002F 百万 token",[58,705,706,709],{},[74,707,708],{},"Output（含思维链）",[74,710,711],{},"¥4 \u002F 百万 token",[35,713,714,715,718],{},"注意：R1 的 Output 价格高于 V3（¥2\u002FM），因为思维链 token 也计入 Output。实际使用中，思维链通常占 Output 的 50-80%，所以",[282,716,717],{},"实际成本约为 V3 的 3-4 倍","。",[35,720,721],{},"夜间折扣（00:30-08:30）同样适用，再 -50%。复杂数学批量任务定时跑夜间。",[31,723,725],{"id":724},"r1-vs-v3-怎么选","R1 vs V3 怎么选",[52,727,728,741],{},[55,729,730],{},[58,731,732,735,738],{},[61,733,734],{},"维度",[61,736,737],{},"R1（推理）",[61,739,740],{},"V3（通用）",[69,742,743,754,764,773,784,795,806],{},[58,744,745,748,751],{},[74,746,747],{},"数学\u002F逻辑",[74,749,750],{},"★★★★★",[74,752,753],{},"★★★☆☆",[58,755,756,759,762],{},[74,757,758],{},"编程",[74,760,761],{},"★★★★☆",[74,763,761],{},[58,765,766,769,771],{},[74,767,768],{},"日常对话",[74,770,753],{},[74,772,761],{},[58,774,775,778,781],{},[74,776,777],{},"速度",[74,779,780],{},"慢（需推理）",[74,782,783],{},"快",[58,785,786,789,792],{},[74,787,788],{},"实际成本",[74,790,791],{},"¥4\u002FM Output",[74,793,794],{},"¥2\u002FM Output",[58,796,797,800,803],{},[74,798,799],{},"多轮对话",[74,801,802],{},"麻烦（思维链要剥）",[74,804,805],{},"简单",[58,807,808,811,814],{},[74,809,810],{},"工具调用",[74,812,813],{},"❌ 不支持",[74,815,816],{},"✅",[35,818,819,606],{},[282,820,821],{},"建议",[276,823,824,827,830],{},[279,825,826],{},"数学 \u002F 推理 \u002F 算法 → R1",[279,828,829],{},"编程 \u002F 对话 \u002F 批量 → V3",[279,831,832],{},"Agent 工具调用 → V3（R1 不支持 function calling）",[31,834,836],{"id":835},"r1-vs-openai-o-series-claude-opus-thinking","R1 vs OpenAI o-series \u002F Claude Opus thinking",[52,838,839,854],{},[55,840,841],{},[58,842,843,845,848,851],{},[61,844,734],{},[61,846,847],{},"R1",[61,849,850],{},"OpenAI o-series（已并入 GPT-5）",[61,852,853],{},"Claude Opus 4 thinking",[69,855,856,868,880,894,905],{},[58,857,858,860,863,866],{},[74,859,140],{},[74,861,862],{},"✅ 完整",[74,864,865],{},"部分（summary）",[74,867,862],{},[58,869,870,872,875,878],{},[74,871,306],{},[74,873,874],{},"✅ MIT",[74,876,877],{},"❌",[74,879,877],{},[58,881,882,885,888,891],{},[74,883,884],{},"价格 Output",[74,886,887],{},"¥4\u002FM",[74,889,890],{},"$10-$60\u002FM",[74,892,893],{},"$75\u002FM",[58,895,896,899,901,903],{},[74,897,898],{},"数学（AIME）",[74,900,20],{},[74,902,82],{},[74,904,101],{},[58,906,907,909,911,913],{},[74,908,810],{},[74,910,877],{},[74,912,816],{},[74,914,816],{},[35,916,917,918,921],{},"R1 的核心定位：",[282,919,920],{},"开源 + 思维链可见 + 价格极低","——研究、教育、蒸馏小模型的首选。",[31,923,924],{"id":924},"适合场景",[35,926,927],{},"✅ 适合：",[276,929,930,933,936,939,942,945],{},[279,931,932],{},"数学题求解 \u002F 证明",[279,934,935],{},"算法设计 \u002F 复杂逻辑推理",[279,937,938],{},"科学问题分析",[279,940,941],{},"学术研究 \u002F 论文公式推导",[279,943,944],{},"代码调试时的根因分析（让 R1 解释为什么 bug）",[279,946,947,948,953],{},"训练数据生成（用 R1 思维链做 SFT 数据，详见 ",[949,950,952],"a",{"href":951},"\u002Fwiki\u002Flora.html","LoRA","）",[35,955,956],{},"❌ 不适合：",[276,958,959,962,965,968],{},[279,960,961],{},"日常对话 \u002F 客服（太慢太贵）",[279,963,964],{},"Agent 工具调用（不支持）",[279,966,967],{},"实时聊天（首 token 等很久）",[279,969,970],{},"简单分类 \u002F 抽取（V3 \u002F Haiku 更划算）",[31,972,973],{"id":973},"避坑清单",[276,975,976,982,988,994,1000],{},[279,977,978,981],{},[282,979,980],{},"不支持 function calling","：R1 不能直接做 Agent 的工具调用层，只能做\"先推理再交给 V3 \u002F GPT 执行\"。",[279,983,984,987],{},[282,985,986],{},"不要传 temperature","：R1 不支持采样参数，传了被忽略，不要从 V3 代码硬迁过来。",[279,989,990,993],{},[282,991,992],{},"思维链不可缓存","：思维链每次重新生成，prompt cache 不能复用思维链——这是 R1 比 V3 贵的根本原因。",[279,995,996,999],{},[282,997,998],{},"multi-turn 把思维链剥掉","：上面已强调，最常见的踩坑。",[279,1001,1002,1005],{},[282,1003,1004],{},"本地跑选蒸馏版","：满血 R1 部署门槛极高（同 V3），普通人用 R1-Distill-7B \u002F 14B 走 Ollama 即可体验。",[31,1007,1008],{"id":1008},"延伸阅读",[276,1010,1011,1017,1024,1031],{},[279,1012,1013,1014],{},"通用兄弟：",[949,1015,106],{"href":1016},"\u002Fmodels\u002Fdeepseek-v3.html",[279,1018,1019,1020],{},"推理模型概念：",[949,1021,1023],{"href":1022},"\u002Fwiki\u002Ftemperature-top-p.html#%E6%8E%A8%E7%90%86%E6%A8%A1%E5%9E%8B%E4%B8%BA%E4%BB%80%E4%B9%88%E4%B8%8D%E5%BB%BA%E8%AE%AE%E6%94%B9-temperature","Temperature 与 Top-P",[279,1025,1026,1027],{},"思维链与 Prompt：",[949,1028,1030],{"href":1029},"\u002Fwiki\u002Fprompt-engineering.html","Prompt Engineering",[279,1032,1033,1034,1037,1038],{},"与同档对比：",[949,1035,76],{"href":1036},"\u002Fmodels\u002Fgpt-5.html"," \u002F ",[949,1039,95],{"href":1040},"\u002Fmodels\u002Fclaude-opus-4.html",[1042,1043,1044],"style",{},"html pre.shiki code .sVt8B, html code.shiki .sVt8B{--shiki-default:#24292E;--shiki-dark:#E1E4E8}html pre.shiki code .szBVR, html code.shiki .szBVR{--shiki-default:#D73A49;--shiki-dark:#F97583}html pre.shiki code .s4XuR, html code.shiki .s4XuR{--shiki-default:#E36209;--shiki-dark:#FFAB70}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .sJ8bj, html code.shiki .sJ8bj{--shiki-default:#6A737D;--shiki-dark:#6A737D}html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":150,"searchDepth":192,"depth":192,"links":1046},[1047,1048,1053,1056,1057,1058,1059,1060,1061],{"id":33,"depth":172,"text":33},{"id":43,"depth":172,"text":43,"children":1049},[1050,1051,1052],{"id":47,"depth":192,"text":47},{"id":140,"depth":192,"text":140},{"id":306,"depth":192,"text":306},{"id":369,"depth":172,"text":370,"children":1054},[1055],{"id":592,"depth":192,"text":593},{"id":673,"depth":172,"text":673},{"id":724,"depth":172,"text":725},{"id":835,"depth":172,"text":836},{"id":924,"depth":172,"text":924},{"id":973,"depth":172,"text":973},{"id":1008,"depth":172,"text":1008},"reasoning",128000,"深度求索 DeepSeek-R1 开源推理大模型，完整暴露思维链（Chain of Thought）推理过程，数学与代码推理能力对标 GPT-5\u002Fo3，API 输入仅 ¥1\u002FM 是 OpenAI 同级的 1\u002F30，国内可直连且权重开放，支持私有部署。","md",32768,{},"\u002Fmodels\u002Fdeepseek-r1","Input ¥1\u002FM (缓存 ¥0.1\u002FM) · Output ¥4\u002FM（含思维链）","2026-06-21",[1072,1073],"coding\u002Fapi\u002Fopenrouter","coding\u002Flocal\u002Follama","2025-01-20",{"title":11,"description":1064},"deepseek-r1","models\u002Fdeepseek-r1",[1079,1080,1081,1082,1083],"开源推理模型，思维链完全可见","数学推理能力接近 GPT-5","价格极低，推理模型中性价比最高","国内直连，响应快","可自行部署（开源权重）",[1085,1086,1087,938],"数学竞赛 \u002F 证明题","复杂逻辑推理","算法设计","深度求索","DeepSeek",[1091,1092,1093,1094],"思维链 token 也计费，实际成本高于 V3","非推理任务不如 V3（速度更慢）","128K 上下文","输出含思维链，需额外解析","QPoX4cd8yNBbcGzFm19_QrMGr08pcUAycWyN8TOyN9I",1782316489330]