商业产业 Hacker News AI 2026-10-01T00:07 正文 ok 650 words

Gemini 4 Argon:Google 重返智能水平领先实验室前三

英文标题:🔶 Gemini 4 Argon: Google is back among the top three labs in intelligence achieved

Google DeepMind 推出的 Gemini 4 Argon 在 Artificial Analysis Intelligence Index 上获得 53 分,与 GPT-6 Astra(max)持平,并领先 GPT-6.1 Sol(max)1 分。该模型目前以 50% 的首发折扣计价,每项 Intelligence Index 任务成本为 1.99 美元,约为 GPT-6 Astra 的 60%,但促销结束后将升至 3.98 美元。Gemini 4 Argon 在 Agent 能力和幻觉率方面明显提升,在 AutomationBench-AA 上排名第一,且在领先模型中拥有最低的 15% 幻觉率。该模型目前仅向部分用户逐步开放,尚未公开发布,开发者还可通过新的 Gemini API Long Decode Continuation 功能支持最长 100 万输出 token 的推理过程。
打开原文相关 2 · 机会 0
media297.1 KB media498.8 KB media179.2 KB media293.9 KB media252.5 KB media423.9 KB media1.1 MB media1.8 MB media10.5 KB media39.1 KB media6.8 KB media2.5 MB

中文

Gemini 4 Argon:Google 重返智能水平领先实验室前三

查看模型页面

Google 新推出的 Gemini 4 Argon 在 Artificial Analysis Intelligence Index 上与 GPT-6 Astra 持平,并凭借折扣价格将单任务成本降至其 60%

Gemini 4 Argon 是 Google DeepMind 7 个多月以来首个定位高于 Flash 系列的专有模型。该模型在高推理强度设置下(可用的最高设置)于 Artificial Analysis Intelligence Index 获得 53 分,与 GPT-6 Astra(max,53 分)持平,并比 GPT-6.1 Sol(max,52 分)高 1 分;其提升主要来自更低的幻觉率和更强的 Agent 能力。

在当前 50% 的价格折扣以及提升至 95% 的缓存折扣下,Gemini 4 Argon 每项 Intelligence Index 任务的成本为 1.99 美元,约为 GPT-6 Astra(max)的 60%,但为 GPT-6.1 Sol(max)的 2.7 倍。折扣结束后,这一成本将升至 3.98 美元,约为 GPT-6 Astra(max)的 1.2 倍。

Gemini 4 Argon 目前正向部分用户逐步推出,尚未公开可用。50% 折扣为初始促销活动,Google 尚未确认促销结束日期。

Gemini 4 Argon 在高推理强度下的关键基准测试结果:

➤ **Google 重返智能水平领先实验室前三:**Gemini 4 Argon(high)在 Artificial Analysis Intelligence Index 上获得 53 分,与 GPT-6 Astra(max,53 分)持平,并比 GPT-6.1 Sol(max,52 分)高 1 分。这比 Google 此前的非 Flash 模型 Gemini 3.1 Pro Preview(30 分)高 23 分,也比 Gemini 3.8 Flash(high)高 12 分。

➤ **50% 的首发折扣使 Gemini 4 Argon 在单任务成本上具备竞争力:**按当前折扣价格,Gemini 4 Argon(high)每项 Intelligence Index 任务成本为 1.99 美元;在相当的智能水平下,这相当于 GPT-6 Astra(max,每项 3.26 美元)成本的 60%。这种成本效率来自更低的 token 价格,而非更少的 token 使用量:Gemini 4 Argon 每项任务平均输出 6.2 万个 token,而 GPT-6 Astra(max)为 2.7 万个 token。Google 尚未确认促销结束日期,但按标准定价,单任务成本将升至 3.98 美元。

**➤ 更强的 Agent 表现:**Gemini 模型历来在这一领域相对较弱,而 Gemini 4 Argon 在多项 Agent 评测中表现出提升。它在 AutomationBench-AA 上以 78% 排名第一,比 Claude Sonnet 5.5(max,71%)高 7 个百分点。在 Terminal Bench 4 上,Gemini 4 Argon 达到 57%,较 Gemini 3.1 Pro Preview 提升 53 个百分点,仅落后于 Claude Sonnet 5.5(max,64%)、Claude Opus 5.5(max,60%)和 GPT-6 Astra(59%)。在 AA-Briefcase 上,它达到 1494 Elo。这主要由 65% 的 rubric 通过率推动,这是我们记录过的最高水平,但其 Analytical Quality(1576 Elo)和 Presentation Quality(1308 Elo)相对较低。

**➤ 领先模型中最低的幻觉率:**在 AA-Omniscience 上,Gemini 4 Argon 的幻觉率为 15%,是所有 Intelligence Index 得分达到 45 分以上模型中最低的;相比之下,GPT-6 Astra(max)为 51%,GPT-6.1 Sol(max)为 54%。这意味着 Argon 更可能承认自己不知道答案,而不是进行错误猜测。在准确率方面,Gemini 4 Argon 得分为 50%,比 Gemini 3.1 Pro Preview 低 5 个百分点,也比 GPT-6 Astra(max,63%)低 13 个百分点。尽管准确率略低,其 AA-Omniscience 总分为 42,仍与 GPT-6 Astra(43)和 GPT-6.1 Sol(42)处于相近水平。

关键模型细节:

➤ **上下文窗口:**100 万 token

➤ **多模态能力:**支持文本、图像、视频和语音输入,文本输出

➤ **定价:**标准价格为每 100 万 input/output token 4 美元/20 美元,目前至少一个月内提供 50% 折扣,降至 2 美元/10 美元。缓存输入 token 可享受 95% 折扣(按折扣价计算,每 100 万 token 为 0.10 美元),高于 Gemini 3.8 Flash 的 90% 折扣。

**➤ 长解码续写(Long Decode Continuation):**我们使用 Long Decode Continuation 测试了 Gemini 4 Argon。这是一项新的 Gemini API 功能,可暂停长响应,并在后续调用中恢复响应。该功能使推理过程可运行至最多 100 万个输出 token,而不会发生请求超时。

如需进一步分析,请参阅 https://artificialanalysis.ai/models/gemini-4-argon

English / 原文

Gemini 4 Argon: Google is back as one of the top three labs in intelligence achieved

See model page

Google’s new Gemini 4 Argon equals GPT-6 Astra on the Artificial Analysis Intelligence Index at 60% of the Cost per Task with discounted prices

Gemini 4 Argon is Google DeepMind’s first proprietary model above the Flash class in over 7 months. With high reasoning (the highest available), it scores 53 on the Artificial Analysis Intelligence Index, matching GPT-6 Astra (max, 53) and 1 point ahead of GPT-6.1 Sol (max, 52), with gains driven by lower hallucinations and stronger agentic capabilities.

At its current 50% pricing discount and with cache discounts increased to 95%, Gemini 4 Argon costs $1.99 per Intelligence Index task, 60% of GPT-6 Astra (max), but 2.7x GPT-6.1 Sol (max). After the discount ends, this will rise to $3.98 (~1.2x GPT-6 Astra (max)).

Gemini 4 Argon is currently being rolled out to selected users and is not publicly available. The 50% discount is an initial promotion. Google has not yet confirmed the promotion end date

Key benchmarking results for Gemini 4 Argon with high reasoning:

➤ Google returns as one of the top three labs on intelligence: Gemini 4 Argon (high) scores 53 on the Artificial Analysis Intelligence Index, matching GPT-6 Astra (max, 53) and 1 point ahead of GPT-6.1 Sol (max, 52). This is 23 points above Google’s previous non-Flash model, Gemini 3.1 Pro Preview (30) and 12 points ahead of Gemini 3.8 Flash (high)

➤ Launch discounts of 50% make Gemini 4 Argon competitive on Cost per Task: At current discounted pricing, Gemini 4 Argon (high) costs $1.99 per Intelligence Index task, 60% of GPT-6 Astra (max, $3.26) for a comparable level of intelligence. This cost efficiency is driven by lower token prices, rather than reduced token use, with Gemini 4 Argon averaging 62k output tokens per task, compared with 27k for GPT-6 Astra (max). Google has not yet confirmed the promotion end date, but on standard pricing, Cost per Task will increase to $3.98

➤ Stronger agentic performance: Historically a weaker area for Gemini models, Gemini 4 Argon shows improvements across agentic evaluations. It ranks #1 on AutomationBench-AA at 78%, 7 points ahead of Claude Sonnet 5.5 (max, 71%). On Terminal Bench 4, Gemini 4 Argon achieves 57%, a + 53 point improvement from Gemini 3.1 Pro Preview, only behind Claude Sonnet 5.5 (max, 64%), Claude Opus 5.5 (max, 60%) and GPT-6 Astra (59%). On AA-Briefcase, it reaches 1494 Elo. This is driven by a 65% rubric pass rate, the highest we have recorded, but lower Analytical Quality (1576 Elo) and Presentation Quality (1308 Elo)

➤ Lowest hallucination rate among leading models: On AA-Omniscience, Gemini 4 Argon has a 15% hallucination rate, the lowest of any model scoring 45+ on the Intelligence Index, compared with 51% for GPT-6 Astra (max) and 54% for GPT-6.1 Sol (max). This means Argon is much more likely to acknowledge when it does not know an answer rather than guess incorrectly. On accuracy, Gemini 4 Argon scores 50%, a 5 point decrease from Gemini 3.1 Pro Preview, and 13 points below GPT-6 Astra (max, 63%). With this slightly lower accuracy, its overall AA-Omniscience score of 42 remains in line with GPT-6 Astra (43) and GPT-6.1 Sol (42)

Key model details:

➤ Context Window: 1M tokens

➤ Multimodality: Text, image, video, and speech input, with text output

➤ Pricing: $4/$20 per 1M input/output tokens at standard pricing, currently discounted 50% to $2/$10 for at least one month. Cached input tokens receive a 95% discount ($0.10 per 1M at discounted pricing), up from 90% on Gemini 3.8 Flash

➤ Long Decode Continuation: We tested Gemini 4 Argon with Long Decode Continuation, a new Gemini API feature that pauses long responses and resumes them across follow-up calls. This lets reasoning run up to 1M output tokens without request timeouts

For further analysis, see https://artificialanalysis.ai/models/gemini-4-argon