OpenAI published first benchmark results for its custom AI inference chip Jalapeño, co-developed with Broadcom, showing 1.5 to 1.9 times greater peak performance per watt than Nvidia Blackwell systems…
OpenAI announced its first custom inference chip, Jalapeño, on August 25, 2026, claiming it outperforms Nvidia's GB200 and GB300 accelerators by 1.5 to 1.9 times more AI work per watt and 1.7 to 3.6 t…
OpenAI reported up to 3.6x lower end-to-end latency for its Jalapeno inference chip compared with Nvidia GB200 and GB300 systems in company-run engineering tests, with production deployment planned by…
OpenAI reported Tuesday that its custom inference chip, Jalapeno, delivered between 1.5 and 1.9 times more AI work per watt at peak throughput than Nvidia GB200 and GB300 systems across three public m…
OpenAI has introduced Jalapeño, its first custom inference chip, developed with Broadcom and Celestica, to improve throughput, latency, and energy efficiency for serving large language models. The chi…
OpenAI said its Broadcom-built Jalapeño inference chip delivers 1.5 to 1.9 times more AI work per watt than Nvidia's GB200 and GB300 systems on tested models, with 1.7 to 3.6 times lower end-to-end la…
OpenAI has disclosed first measured results for its custom inference chip, Jalapeño, showing a 1.5 to 1.9 times increase in AI work per watt at peak throughput compared to leading commercial systems, …
Multiverse Computing's new Quantization-Aware Healing (QAH) method, applied to a GPT-OSS 120B model compressed to 60B parameters and quantized to MXFP4, produces a 4-bit model that outperforms its ful…
A developer built a capture-the-flag arena where language models attack and defend containers, and trained a local Qwen2.5-3B-Instruct bot with an MLX LoRA adapter on game replays. The bot outperforme…
Snyk, a security company, has introduced a model risk evaluation framework that measures the security risk of large language models by running them as agents in dynamic environments against adaptive a…
Three independent benchmarks show that expensive AI defaults are rarely optimal: medium reasoning effort captures nearly all of Claude Opus 5's bug-fix gains, a three-model open-weight jury (GPT-OSS 1…
A study on arXiv (2608.00004v1) found that three cheap open-weight models — GPT-OSS 120B, DeepSeek-V4 Flash, and Gemma-4 31B — agree with human pass/fail decisions on natural-language mathematical pro…
AMD's Ryzen AI Max Plus 395 chip, featuring 128GB unified LPDDR5X memory and a Radeon 8060S GPU, enables local machines to run 100B+ parameter large language models like GPT-OSS 120B without a discret…
On a 128GB unified memory system using AMD's Ryzen AI Max Plus 395 chip, a dense 9B model with a speculative drafter ran at just under 40 tokens per second, a 35B mixture-of-experts Qwen3.6 model with…
Researchers have developed AutoWorldBuilder, a multi-agent collaborative system that uses Large Language Models to automate fictional worldbuilding, achieving a 95.0% success rate across 20 diverse ta…
Independent benchmarks of NVIDIA's DGX Spark desktop through mid-2026 show Qwen 3.5 27B as the most consistent all-rounder on the easy Ollama path, while GPT-OSS 120B pushes nearly 4x the throughput b…
Sierra Research introduces the Context Engine and meta-distillation to address specification failure in enterprise agents, improving retrieval and action-check pass rates on benchmarks like τ³-bench, …
Neuralwatt launched the first AI inference API with energy-based pricing, charging per kilowatt-hour instead of per token to provide transparency into power consumption and cost. The platform offers r…
A developer abandoned attempts to make an AI website builder generate proper desktop layouts through prompt engineering after two weeks of failed iterations. The engineer found that large language mod…