分析公司Gartner指出，以营利为目的的模型开发商将只能从代理式AI工作流所涉步骤中的10%获取收入，而其中许多开发商正投入数十亿美元建设AI基础设施。该公司高级总监马克·佩蒂和集团副总裁雷蒙德·帕凯上周在新文件《开放权重模型将运行90%的企业代理式工作流步骤》中写道：“代理工作流步骤中有九成是常规性的，开放权重模型能以最便宜闭源模型四分之一的价格、在同等水平上处理这些步骤。”
For-profit model-makers, many of which are spending billions on AI infrastructure, will be chasing revenue from just ten percent of steps involved in agentic AI workflows, according to analyst firm Gartner. “Nine in 10 agent workflow steps are routine, and open-weight models handle them at parity for a quarter of the price of the cheapest closed model,” the firm’s senior director Mark Petty and group vice president Raymond Paquet wrote last week in new document titled “Open-Weight Models Will Run 90% of Enterprise Agentic Workflow Steps.”
两人观察到，“典型的运营工作流涉及数百次常规执行，例如分类、工具调用、传递、验证和核对数据”，随后提出开放权重模型“可以为每一个需要前沿推理的步骤处理数百个这样的常规步骤”。两人补充说：“在相同得分下，与最便宜的闭源模型相比，开放模型的成本只有四分之一。如果每一步都使用前沿模型，开放模型的成本则低至二十分之一。”
“Typical operational workflows involve hundreds of routine executions, such as classification, tool calling, passing, validating and reconciling data,” the pair observe, before suggesting open-weight models “can handle hundreds of these routine steps for every one that needs frontier reasoning.”“Against the cheapest closed model at the same score, the open model costs a quarter as much. Against a frontier model run on every step, it costs as little as a twentieth,” the pair add.
这些是最理想情况下的数字，因为分析师承认，“当前开放模型可能更啰嗦、重试更频繁，因此与前沿模型相比，每项任务节省的成本更接近五到六倍。”这仍然是可观的节省，但分析师将其视为转向开放模型带来的可喜副产品，而非目的本身。两人写道：“多年的工作流历史是收购方在为SaaS公司支付对价时的重要部分，而每一个发送到前沿API的常规步骤，都会把这部分历史中的一片移到你的边界之外。开放权重能让你保持控制权。”
Those are the best-case numbers, because the analysts admit “Current open models can be more verbose and retry more often, so the saving per task against a frontier model is closer to five or six times.”That’s still a lovely saving, but the analysts rate it as the welcome by-product of moving to open models rather than an end in itself. “Years of workflow history are a large part of what an acquirer pays for in a SaaS company and every routine step sent to a frontier API moves a slice of it outside your perimeter,” the pair wrote.
“成本节约是一个额外收益。”因此，两位分析师提出以下战略规划假设：“到2030年，企业运营代理工作流的90%步骤将在开源权重模型上执行。”他们认为企业将主要采用领域特定的开源权重模型，因为这些模型在“少数高度稳定的步骤每月运行数百万次的场景”中表现良好。然而，大多数模型已经成为商品。Harness（即运行代理并管理代理的工具）是AI产品价值的来源，因为其角色之一是将工作路由到最合适的模型完成，而用户几乎感受不到任何影响。
“Open-weights preserve your control. The cost saving is a bonus.”The two analysts therefore offer the following strategic planning assumption: “By 2030, 90% of enterprise operational agent workflow steps will execute on open-weight models.”They argue that enterprises will mostly adopt domain-specific open-weight models, because they do well “in scenarios where a small number of highly stable steps run millions of times a month.”Most models, however, are already commodities. Harnesses, the tools that run agents and manage agents, are the source of value in AI products, because one of their roles is to route work to the most appropriate model for a job, without users noticing any impact.
一个优秀的harness能够节约成本、施加强有力的安全防护，并管理代理舰队。功能较弱的harness则会让用户需要自行处理更多事务。论文指出，前沿模型将继续作为研发的引擎，实际上属于资本支出。前沿模型完成重活后，开源权重模型则转变为运营支出。
A good harness saves money, imposes strong guardrails, and manages agent fleets. Less powerful harnesses leave users with more to do. The paper suggests frontier models will continue to have a role as the engine of research and development and are effectively capital expense. Open-weight models become an operational expense after frontier models do the heavy lifting. ®