grok-4.3

Grok 4.3, released in May 2026, is an inference model launched by xAI before xAI was merged into SpaceX and renamed SpaceXAI. It is positioned as a general flagship for complex tasks such as advanced logic, mathematics, scientific analysis and multi-step investigation. Compared with the previous generation Grok 4.20, the architecture of Grok 4.3 has been improved, the context window has been expanded to 1 million Tokens, and the knowledge deadline is December 2025. The model supports text and image input, text output, and has the ability to function calls, structured output, and cache input tokens.

Context
1000000 tokens
Max output
128000 tokens
Input
llm image
Output
llm

Pricing

Input: $1.25 / 1M tokens

Output: $2.5 / 1M tokens

Cached input: $0.2 / 1M tokens

Endpoints

  • openai
  • openai-response

More models

  • vertex/gemini-3.5-flash-lite
  • qwen3.7-max

    Qwen3.7-Max 于 2026 年 5 月 20 日发布,是千问 3.7 系列的旗舰版本,定位为“面向智能体时代的新一代旗舰模型”。模型拥有超过 1 万亿参数及 100 万 Token 上下文窗口,核心优势在于智能体能力的广度与深度,在编程、办公生产力和长周期自主执行方面均有突出表现。在 Terminal-Bench 2.0 Terminus 上得分 72.3,SWE-Pro 得分 60.4,多项指标刷新国内 SOTA。其标志性能力是支持 35 小时连续自主运行、完成超过 1,100 次工具调用,并在全自主 kernel 优化任务中达成 10.0 倍几何平均加速比。模型兼容 Claude Code、OpenClaw 等主流智能体框架,是构建复杂 Agent 系统的可靠基座。

  • claude-opus-5

    Claude Opus 5 是面向深度推理、复杂编码、研究分析与高难度创意工作的高级模型。它支持自适应思考,并默认开启思考过程;可通过 effort 调整任务投入程度,适合长周期智能体、复杂软件工程、资料研究、文档生成和多步骤工具编排。该模型支持大上下文、提示缓存和较长输出,但 max_tokens 同时约束思考与最终答案,因此应重新评估旧工作流的输出上限。生产部署前应重点验证延迟、成本、工具调用稳定性和任务边界。

  • gpt-6-luna