OpenAI Unveils 'Jalapeño' Custom Inference Chip with Broadcom
OpenAI has officially revealed its first custom-designed silicon, an ASIC codenamed 'Jalapeño,' developed in collaboration with Broadcom. Optimized specifically for large language model (LLM) inference, the chip reportedly delivers a 50% reduction in operational costs compared to current industry standards like Nvidia’s Blackwell GPUs. Developed in a record-breaking nine-month cycle, Jalapeño marks a decisive move by OpenAI to achieve hardware independence and manage the soaring energy demands of its frontier models.
AI pioneer Andrej Karpathy has joined Anthropic's pre-training team to lead an initiative using Claude to improve its own training efficiency, a strategy termed 'model-accelerating-model.'
OpenAI launched GPT-5.5-Cyber, a specialized defender model that outscored Anthropic’s Mythos 5 on the CyberGym benchmark, achieving an 85.6% success rate in identifying vulnerabilities.
Anthropic has filed a formal complaint against Alibaba, alleging the company accessed Claude models through over 28 million unauthorized exchanges via fraudulent accounts.
The Trump administration is reportedly pressuring Meta to submit its frontier models for voluntary federal review, as it remains the only major U.S. lab yet to sign the agreement.
Cambridge experts released a report warning that the current generation of frontier AI models could significantly empower malicious actors in launching biological and chemical threats if safety protocols are not standardized.
Did you know?
OpenAI’s Jalapeño chip achieved 'tape-out'—the final stage of the design process before manufacturing—in just nine months, making it one of the fastest development cycles for high-performance AI hardware in history.
Quiz
Which AI model recently set a new record on the GPQA Diamond benchmark (graduate-level science reasoning) with a score of 82.4%?