OpenAI号发射GPT-5.3-Codex-Spark号,这是切列布拉斯芯片的快速、轻量级编码模型,提供实时发展速度和减少对Nvidia的依赖。 OpenAI launches GPT-5.3-Codex-Spark, a fast, lightweight coding model for Cerebras chips, offering real-time development speed and reduced reliance on Nvidia.

OpenAI已发布GPT-5.3-Codex-Spark(GPT-5.3-Codex-Spark),这是一个快速、轻量级的编码模型,优化用于切列布拉斯系统(Cerebras Systems)的裂谷级芯片,每秒交付1,000个标记,比以往的模型输出速度快25%。 OpenAI has released GPT-5.3-Codex-Spark, a fast, lightweight coding model optimized for Cerebras Systems’ wafer-scale chips, delivering over 1,000 tokens per second and 25% faster output than prior models. 它设计用于实时开发任务,如调试和文档,使用早期版本的一半符号,在终端-Bench 2.0 上分数77.3%,仅用44GB的内存运行。 Designed for real-time development tasks like debugging and documentation, it uses half the tokens of earlier versions, scores 77.3% on Terminal-Bench 2.0, and runs with just 44 GB of memory. 它标志着OpenAI的第一个非Nvidia硬件主要生产模型, 这是与Cerebras建立100亿美元合作伙伴关系的一部分。 Available now to ChatGPT Plus and Codex Pro users, it marks OpenAI’s first major production model on non-Nvidia hardware, part of a $10 billion partnership with Cerebras. 该模型虽然牺牲了广泛的速度能力,但标志着向多样化的人工智能硬件的战略转变,减少了对Nvidia的依赖,并强调开发商的低长推论。 While it sacrifices broad capabilities for speed, the model signals a strategic shift toward diversified AI hardware, reducing reliance on Nvidia and emphasizing low-latency inference for developers.