中国开源模型在 AI 编码领域挑战 Anthropic

China’s Open-SourceModels ChallengeAnthropic in AI Coding 中国开源模型在AI编码领域挑战Anthropic Chinese companies arelaunching open-source AI models built to power codingassistants as cheaper alternatives to thoseavailable fromOpenAI,AnthropicandGoogle. 中国公司正在推出开源AI模型，旨在为编码助⼿提供动⼒，作为OpenAI、Anthropic和⾕歌等公司产品的更经济替代⽅案。 Late last month, Shanghai-based AI startupMiniMaxlaunchedM2, a new open-sourcelarge language model developed primarilyfor coding and AI agents. In apost on X,MiniMax noted that the price of M2 is just8% of that of Anthropic’sClaude Sonnetmodel. (M2 costs $1.20 per a million outputtokens, compared with $15 for ClaudeSonnet, according to both companies.) 上⽉末，总部位于上海的AI初创公司MiniMax发布了M2，这是⼀款主要为编码和AI代理开发的新型开源⼤型语⾔模型。MiniMax在X上发帖指出，M2的价格仅为Anthropic的Claude Sonnet模型的8%。（根据两家公司的数据，M2每百万输出token的成本为1.20美元，⽽Claude Sonnet为15美元。） M2 has already gotten popular. Startupsthat develop coding agents, such asCline,Kilo CodeandRoo Code, have made M2available to their users. M2 is currently thefourth most popular model based on tokenusage onOpenRouter, which offershundreds of models to developers. M2trails onlyxAI,Anthropic and Google onthe site. M2已经⼴受欢迎。开发编码代理的初创公司，如Cline、Kilo Code和Roo Code，都已向其⽤户提供M2。在向开发者提供数百种模型的OpenRouter平台上，M2⽬前是基于token使⽤量的第四⼤最受欢迎模型。在该⽹站上，M2仅次于xAI、Anthropic和Google。 The low cost isn’t M2’s only selling point. Itranks No. 4 for web development,according toLMArena, which compares AImodels on user votes. And it ranks No. 5 interms of intelligence on another AI modelleaderboard, compiled by benchmarkingcompany Artificial Analysis. 低成本并⾮M2的唯⼀卖点。根据LMArena（⼀个通过⽤户投票⽐较AI模型的平台），M2在⽹络开发⽅⾯排名第四。⽽在由基准测试公司Artificial Analysis编制的另⼀个AI模型排⾏榜上，M2在智能⽅⾯排名第五。 MiniMax’s M2 is part of a wave of newChinese open-source models that are tryingto appeal to developers in the U.S. andelsewhere who are looking for decent butmore affordable AI models for coding.DeepSeekbuilt a reputation as a providerof strong coding models. More recentlyother Chinese startups, such as MiniMax,Moonshot AIandZhipu, have beenreleasing new AI models that show strongcoding capabilities. MiniMax的M2是中国开源模型浪潮中的⼀员，这些模型正试图吸引美国及其他地区寻求价格合理且性能良好的编码AI模型的开发者。DeepSeek在提供强⼤的编码模型⽅⾯建⽴了声誉。最近，MiniMax、Moonshot AI和Zhipu等其他中国初创公司也发布了展现出强⼤编码能⼒的新AI模型。 They tend to be far cheaper than U.S.models, raising the potential for a pricewar that could hurt American AI modelmakers such as Anthropic that owe asignificant portion of their revenue growthto powering AI agents. 它们往往⽐美国模型便宜得多，这可能引发⼀场价格战，从⽽损害Anthropic等美国AI模型制造商的利益。这些制造商的收⼊增⻓很⼤⼀部分都依赖于为AI代理提供⽀持。 For the Chinese startups, it is critical theyfind customers outside China who pay toaccess the AI models through applicationprogramming interfaces, no matter howlow the prices are. That’s because it isdifficult for AI companies in China togenerate revenue from domestic customerswhere price competition is fierce andbusiness customers are reluctant to pay forsubscriptions. 对于中国的初创公司来说，⾄关重要的是，⽆论价格多低，他们都要找到中国以外的客户，通过应⽤程序编程接⼝（API）付费访问⼈⼯智能模型。这是因为中国的AI公司很难从国内客户那⾥获得收⼊，国内价格竞争激烈，企业客户也不愿为订阅付费。 Chinese open-source AI models could facescrutiny from U.S. lawmakers worriedabout the risk they pose to nationalsecurity. And many customers are willingto continue spending extra money on morecapable American models they considerreliable. But as the overall AI codingmarket grows rapidly, the Chinesecompanies are betting that there will besufficient demand for cheaper and goodenough options. 中国开源⼈⼯智能模型可能会受到美国⽴法者的审查，他们担⼼这些模型对国家安全构成⻛险。许多客户也愿意继续花更多的钱购买他们认为可靠的、功能更强⼤的美国模型。但随着整个AI编码市场的迅速增⻓，中国公司认为，对更便宜、⾜够好的选择会有⾜够的需求。 Upgrade to askDeep Research to…升级以让深度研究来……………………