OpenAI's Jalapeño Chip Outperforms Nvidia in Inference Benchmarks
What happened
Chinese AI lab Z.ai has confirmed its GLM-5.3-Flash model is the anonymous 'Ox Alpha' that topped OpenRouter's rankings, doubling DeepSeek's performance. This model, released with open weights, is priced at a discounted $0.045 per task, approximately one-tenth of comparable rivals, and achieved its record performance on domestic chips.
Why it matters
The launch of Z.ai's GLM-5.3-Flash model, offering high performance at a tenth of the cost and running on domestic chips, signals a significant shift in AI accessibility and hardware independence, compelling practitioners to evaluate new, cost-effective models for their deployments.
Topics
- GLM-5.3-Flash
- Large Language Models
- AI Chips
- Cost Efficiency
Articles in this trend
- OpenAI's first custom chip "Jalapeño" reportedly beats Nvidia's Blackwell and Rubin in inference benchmarks — The Decoder
- OpenAI's first AI chip brings the heat — The Rundown AI
- OpenAI says its Jalapeño chip can power faster AI responses than the competition — The Verge
- OpenAI Says Its Jalapeño AI Chip Is Better Than Nvidia’s Blackwell — The Information
- OpenAI’s Jalapeño chip is built for fast inference at scale, benchmarks show — TechCrunch
- OpenAI BROKE the Industry Overnight.... — Wes Roth
- Why OpenAI’s Jalapeño Might Sicken Nvidia — The Information
- OpenAI says its AI chip rivals Nvidia in inference — Semafor
- 🔴 LIVE: OpenAI Details Jalapeño Chip | AM Intelligence Orders 9,000 Rubin GPUs | Front Page — AIM Network
- 656: Google vs. Nvidia’s Full Stack, OpenAI’s Spicy Jalapeño, Moderna’s $500K Cancer Bet, Google Maps, I Apparently Like Grok Now, Altman’s AI Bottlenecks, a Tiny AI Scientist, and Metallica — Liberty’s Highlights
- HUGE: OpenAI's Jalapeño Chip Outperforms Nvidia Blackwell by Up to 4x on Inference — AIM Network
- The Ox Alpha mystery ends with Z.ai — The Rundown AI