Fireworks AI 推出 FireRouter with Opus 路由模型
Introducing FireRouter with Opus
Fireworks AI 发布 FireRouter with Opus，这是其首个缓存感知路由器，已在 CLI 上线并首次以独立路由模型形式提供，任何 Fireworks 账号都可将其作为 serverless endpoint 调用。该路由器在 Claude Opus 5.5、GLM 5.3 和 GLM 5.3 Flash 之间按任务适配度、模型成本与缓存命中损失进行路由。经过一个多月内部 A/B 测试，编码任务准确率为单独使用 Opus 的 98.1%，每次会话成本从 15.36 美元降至 6.63 美元，降幅 57%（±19 ppts），缓存命中率为 94.2%，单独使用 Opus 时为 97.8%。
推荐理由
精选AI 评分 69
原文给出内部 A/B 测试的成本与准确率数据，读者可据此判断缓存感知路由在编码任务中的取舍。
主题
正文
当前显示原文，中文全文译文尚未就绪。
Join us for our inaugural conference, Forge 2026
PUBLISHED 9/28/2026
Fireworks Nexus enables engineering teams to drop leading open models in the harnesses they already use and cut spend in half without sacrificing speed or quality. The solution includes FireRouter, the first cache-aware router on the market, which makes a big difference in speed and cost.
Today, we’re introducing FireRouter with Opus , optimized for the Opus family and now available in both our CLI and, for the first time, as a standalone router model. Any Fireworks account can point to it as a serverless endpoint as you would any other model.
After more than a month of internal A/B testing, FireRouter with Opus executes coding tasks at 98.1% of the accuracy for 57% lower cost versus Opus alone.
Today, FireRouter with Opus routes between Claude Opus 5.5, GLM 5.3, and GLM 5.3 Flash Each user turn that hits FireRouter gets evaluated on how well each model in the set is suited for the task. It then estimates the cost of each model handling the task, including the cost of the prompt cache and whether losing it to switch to a different model is worth it. And finally, it routes to the model that best balances quality against cost.
At the moment, a user turn will be routed between Claude Opus 5.5, GLM 5.3, and GLM 5.3 Flash. The model set will change as new models and versions launch.
Our internal coding traffic was on sessions randomly assigned to FireRouter with Opus or to an Opus-only control group, across the same workloads and users.
Cost per session fell 57% (±19 ppts), from $15.36 to $6.63.
To score the accuracy, we verified whether the agent finished the work, whether the right answer was reached, and whether the user had to make any corrections on the next turn.
FireRouter with Opus scored 78.7% of graded turns against 80.2% for Opus-only, or 98.1% of its overall accuracy.
Most of our traffic came from internal coding work, where open models could handle many routine turns. FireRouter’s cache-aware routing weighed the savings from switching models against the cost of losing cache hits. The result was a 94.2% cache hit rate with FireRouter and Opus, compared to 97.8% with Opus alone– a small, deliberate tradeoff that substantially reduced overall cost.
Copy 1 2 fireconnect login fireconnect claude --model firerouter/opus Log in to Fireworks to see estimated savings on your team’s Claude Opus spend.