谷歌表示,其新款Gemini 3.8 Flash模型“更卖力”,但成本可能更高。

qimuai 发布于 阅读:55 一手编译

谷歌表示,其新款Gemini 3.8 Flash模型“更卖力”,但成本可能更高。

内容来源:https://www.theverge.com/ai-artificial-intelligence/988742/google-gemini-3-8-flash

内容总结:

谷歌近日发布新一代轻量级AI模型Gemini 3.8 Flash,距其上一代版本发布仅数周时间。谷歌宣称,该模型在复杂任务处理上“更加卖力”,能够执行更多推理步骤并“迭代式调用工具”。虽然其基础定价与3.7 Flash保持一致(每百万输入tokens收费0.75美元,每百万输出tokens收费3.75美元),但用户的实际使用成本可能有所上升。谷歌方面提醒称,“该模型为追求最优表现,尤其在较高工作量级别下,可能会消耗更多tokens”。若开发者希望控制token使用量,仍可继续选用Gemini 3.7 Flash。

与此同时,谷歌还推出了针对网络安全场景的Gemini 3.8 Flash Cyber版本,并同步启动全新的“Fairwind计划”。该计划仅面向政府及“受信任合作伙伴”,目前已有CrowdStrike、互联网安全中心等650家机构加入。通过该计划,用户可获得3.8 Flash Cyber及谷歌CodeMender智能体服务,后者据称能“自主发现并修复漏洞,保护关键基础设施、公共服务和国家安全”。

模型发布后,业内初步评测反响积极。分析机构Artificial Analysis指出,该模型在同等智能水平下价格极具竞争力,但“尽管单tokens价格未变,由于每项任务平均输出tokens增加约30%,且智能体评测中调用轮次更多,整体成本较3.7 Flash上升约40%”。AI企业Aigora.ai首席执行官John Ennis则将其与Anthropic旗下模型对比,称其“具备Opus 5级代码质量,但成本极低且速度飞快,非常适合用于生成Remotion视频等场景。”

谷歌表示,Gemini 3.8 Flash在软件工程和自主AI智能体领域实现“显著提升”。在DeepSWE v1.1软件工程基准测试中,该模型表现优于前代及其他前沿模型,包括本周刚刚升级并同样主打高性价比的Anthropic Fable 5(后者通过降低缓存数据调用费用来实现成本优化)。此外,新模型在Vals金融智能体V2基准和Harvey法律智能体基准上的表现也超越竞争对手。不过,谷歌同样为该模型内置了针对化学、生物、放射与核(CBRN)及网络攻击等领域的滥用防护措施。

目前,Gemini 3.8 Flash已面向消费者开放,用户可通过谷歌AI Pro或Ultra订阅使用,同时开发者和企业用户也可接入调用。

中文翻译:

谷歌发布了Gemini 3.8 Flash,距离其上一代产品问世仅数周时间。该公司声称,这款新模型比Gemini 3.7 Flash“更卖力”,在复杂任务上会执行更多推理步骤,并且“迭代式地调用工具”。它的初始定价与3.7 Flash相同,即每百万输入token收费0.75美元、每百万输出token收费3.75美元,但最终仍可能让用户花费更多。谷歌提醒说,“该模型可能会使用更多token来最大化性能,尤其是在较高努力程度下。”如果开发者希望将token用量降到最低,可以继续使用Gemini 3.7 Flash。

谷歌称其新款Gemini 3.8 Flash模型“更卖力”,但成本可能更高
Gemini 3.8 Flash Cyber也随谷歌新推出的Fairwind计划一同发布。
Gemini 3.8 Flash Cyber也随谷歌新推出的Fairwind计划一同发布。
Gemini 3.8 Flash发布后,网上很快就出现了一些初步评价。Artificial Analysis重点指出了该模型的定价,称Gemini 3.8 Flash是“我们在这个智能水平上测到的最便宜的模型。尽管每token定价未变,但相比Gemini 3.7 Flash成本仍上升了约40%,这主要是由于每项任务的输出token增加了30%,并且在智能体评测中循环次数更多。”

Aigora.ai首席执行官约翰·恩尼斯将Gemini 3.8 Flash与Anthropic的模型进行了对比,称它提供了“Opus 5级别的编码质量,但成本只是其一小部分,而且速度极快”,并补充说,“这对于制作类似remotion视频这类事情来说将非常棒。”

谷歌表示,Gemini 3.8 Flash在软件工程和自主AI智能体方面带来了“显著改进”。在DeepSWE v1.1软件工程基准测试中,它的表现超越了其前代以及其他前沿模型,包括Anthropic的Fable 5——后者本周早些时候获得了升级,同样承诺以更低价格提供更强性能,做法是降低了使用缓存数据的费用。

谷歌的新模型还在Vals Finance Agent V2基准测试和Harvey法律智能体基准测试中击败了竞争对手。不过,这款新模型也“在化学、生物、放射性和核(CBRN)以及网络攻击等领域内置了防范滥用的保障措施。”

谷歌还随其新的Fairwind计划发布了Gemini 3.8 Flash Cyber,该计划仅限于政府机构和“可信合作伙伴”。该计划目前有650个成员,其中包括CrowdStrike和互联网安全中心(Center for Internet Security),成员可以使用3.8 Flash Cyber以及谷歌的CodeMender智能体——谷歌表示后者能够帮助“自主发现并修复漏洞,保护关键基础设施、公共服务和国家安全。”Gemini 3.8 Flash现已向订阅了谷歌AI Pro或Ultra的消费者提供,同时也可供开发者和企业用户使用。

热门推荐

英文来源:

Google launched Gemini 3.8 Flash, arriving just a few weeks after its predecessor. The company claims the new model “works harder” than Gemini 3.7 Flash by performing more reasoning steps on complex tasks and “calling tools iteratively.” It has the same introductory pricing as 3.7 Flash, $0.75 per million input tokens and $3.75 per million output tokens, but could still end up costing users more. Google warns that “the model might use more tokens to maximize performance, especially at higher effort levels.” Developers can keep using Gemini 3.7 Flash if they want to minimize token usage.
Google says its new Gemini 3.8 Flash model ‘works harder’ but might cost more
Gemini 3.8 Flash Cyber is also launching to Google’s new Fairwind Program.
Gemini 3.8 Flash Cyber is also launching to Google’s new Fairwind Program.
Gemini 3.8 Flash’s launch was followed by some early impressions online. Artificial Analysis highlighted the model’s pricing, saying Gemini 3.8 Flash is “the cheapest we’ve measured at this level of intelligence. This is up ~40% from Gemini 3.7 Flash despite unchanged per-token pricing, driven by a 30% increase in output tokens per task and more turns on agentic evaluations.”
Aigora.ai CEO John Ennis compared Gemini 3.8 Flash to Anthropic’s models, saying it offers “Opus 5 coding quality but at a fraction of the cost and super fast,” adding, “This is going to be so awesome for things like making remotion videos.”
Google says Gemini 3.8 Flash offers “significant improvements” for software engineering and autonomous AI agents. It outperforms its predecessor and other frontier models on the DeepSWE v1.1 software engineering benchmark, including Anthropic’s Fable 5, which got an upgrade earlier this week that also promises stronger performance at a lower price, by cutting the price to use cached data.
Google’s new model also outperformed its competitors on the Vals Finance Agent V2 benchmark and Harvey’s Legal Agent benchmark. However, the new model also “ships with safeguards against misuse in the domains of Chemical, Biological, Radiological, and Nuclear (CBRN) and cyber offense.”
Google also released Gemini 3.8 Flash Cyber alongside its new Fairwind Program, which is limited to governments and “trusted partners.” The program, whose 650 members include CrowdStrike and the Center for Internet Security, offers access to 3.8 Flash Cyber and Google’s CodeMender agent, which Google says can help “autonomously find and fix vulnerabilities, protecting critical infrastructure, public services, and national security.” Gemini 3.8 Flash is available now for consumers with a Google AI Pro or Ultra subscription, as well as developers and enterprise users.
Most Popular

ThevergeAI大爆炸

文章目录


    扫描二维码,在手机上阅读