新的基准测试显示，Qwen3.8的推理能力需要巨大的计算成本，一项编码任务耗时11小时，一个SVG请求消耗了超过22000个token。禁用推理功能可显著减少token使用量和处理时间，这突显了在本地部署中高级推理与效率之间的权衡。与此同时，OpenRouter的估值大幅增长，在八个月内增长了五倍多，达到70多亿美元，为800万开发者提供服务，并支持访问400多个AI模型。 AI
影响 Qwen3.8等模型的先进推理的高成本可能会影响部署策略，而OpenRouter的快速估值增长表明投资者对AI基础设施平台充满信心。
排序理由 该集群涵盖了对前沿模型的重大成本分析以及AI平台的重大估值跃升。
在 Mastodon — fosstodon.org 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
The reasoning setting carries real costs. One coding task ran 11 hours on Qwen3.8's default before finishing well. Disabling reasoning cuts token use by 85 percent. Users benchmarking this week likely measured maximum expense and reported it as typical performance. https://www. i…
A test SVG request on Qwen3.8-27B took 22,276 reasoning tokens over 21 minutes at default settings, versus 3,715 tokens in 137 seconds with reasoning off. The reasoning cost matters when deploying locally. https://www. implicator.ai/qwen-3-8-27b-xhi gh-reasoning-default/ # ai # i…
OpenRouter's valuation jumped more than five times in eight months, from $1.3 billion in May to over $7 billion now. The platform serves 8 million developers across 400+ AI models. Watch whether integrating payments into the routing layer changes how developers access different A…