grok-4.5

Grok 4.5 was released by SpaceXAI on July 8, 2026. It is the first flagship model after xAI merged with SpaceX and acquired the AI ​​programming tool Cursor. The model is built based on the V9 basic model with 1.5 trillion parameters, and is trained using tens of thousands of NVIDIA GB300 GPUs. Cursor's real programming data is added in the supplementary training phase to focus on strengthening programming and agent capabilities. The context window supports up to 1 million Tokens.

Context
500000 tokens
Max output
128000 tokens
Input
llm image
Output
llm

Pricing

Input: $2 / 1M tokens

Output: $6 / 1M tokens

Cached input: $0.3 / 1M tokens

Endpoints

  • openai
  • openai-response

More models

  • kimi-k3

    kimi-k3 是 Moonshot AI 的 Kimi 系列模型标识，Kimi K3 版本，适合复杂文本理解、问答和长材料处理。它可用于长文档梳理、研究资料问答、代码协作、写作辅助和企业知识检索。实际可用的上下文长度、工具能力、缓存与计费方式应以 Moonshot 或接入云平台的官方模型页为准。使用时应把任务目标、文档范围、引用格式和输出结构写清楚；对合同、财务、医疗等需要高准确性的内容，应保留原文定位、来源核验和人工审批。

  • claude-opus-4-8

    Claude Opus 4.8 是 Opus 4 系列中较新的高能力稳定型号，面向复杂推理、专业编程、长材料分析、智能体协同和高要求创意任务。它适合技术方案设计、跨文档研究、复杂代码库维护、任务分解与关键业务流程辅助。与较早版本相比，使用时更应依赖清晰提示、固定测试集和工具调用约束，而非依赖过度强调“反复检查”的提示语。对关键结论建议保留引用来源、规则校验和人工复核。

  • gemini-3.8-flash-0930
  • claude-opus-5-1009