第一项始终是图钉的来源。总体可信度是各份资料对上方所用开始和结束时间支持程度的加权平均;资料每比最新的一份旧 180 天,权重减半。
Artificial Analysis[3]的发布文章日期为2026年4月30日,开头写道'xAI已发布Grok 4.3';OpenRouter[6]列出x-ai/grok-4.3的创建日期为2026-04-30,AA的模型页面也给出2026年4月30日;测试版自4月17日起已在Grok应用中上线(Codersera[4]),部分媒体将全面发布时间定为5月1日。
The maker's model page for grok-4.3: text and image in, text out, 1,000,000-token context, $1.25 per 1 million input tokens, $0.20 cached and $2.50 output, higher rates past 200K tokens, reasoning effort none to xhigh with low the default, function calling and structured outputs.[1]
AA's model page: released April 30, 2026, 1M context, text and image in, $1.25 and $2.50 per 1 million tokens, about 103 output tokens per second with 21.8 s to first token, 87M output tokens to run the index, now scored 25 on the current index and marked deprecated in favour of newer Grok releases.[1]
Secondary guide with the timeline (beta on 17 April, public API on 30 April), native video input up to 5 minutes at 1080p, always-on reasoning, and the later AA index rebase to 24.88 on v4.3.2 against 25.22 for Grok 4.20.[1]
Independent report: built for developers with a focus on speed, low cost and tool calls (xAI's Eric Jiang), December 2025 knowledge cutoff, native web search, X search, code execution and file search, Excel, PDF and PowerPoint output, $395 to run the AA index against $3,959 for GPT-5.5, and mixed results from Andon Labs and Vals AI.[1]
有遗漏或错误吗?用你自己的话说明:能佐证此图钉的链接、不同的开始或结束日期及理由,或缺失、有误的信息。AI 会对照此图钉的来源进行核实,搜索更好的来源,并添加任何支持你说法的页面。图钉自身的来源仍然最重要。AI 也会查看图片:显示的是别的东西或显示效果差的图片会被移到后面或替换。
原标题: Grok 4.3 Released
| 指标 | Grok 4.3 | Grok 4.20 (0309 v2) |
|---|---|---|
| 智能指数 (v4.0) | 53 | 49 |
| GDPval-AA Elo | 1,500 | 1,179 |
| tau2-Bench Telecom | 98% | 93% |
| IFBench | 81% | 81% |
| 运行该指数的成本 | $395 | $514 |