Microsoft benchmarks RTX Spark laptops at 64GB, not the 128GB it sells
Microsofts RTX Spark laptops are sold on 128GB of unified memory. All three MacBook speed claims were measured on 64GB preproduction units, two by NVIDIA.
Read MoreIndependent AI news, with every claim linked to its primary source
Independent AI news, with every claim linked to its primary source
New models, benchmarks and research from the labs.
Microsofts RTX Spark laptops are sold on 128GB of unified memory. All three MacBook speed claims were measured on 64GB preproduction units, two by NVIDIA.
Read MoreMistral Large 4 is a 1 trillion-parameter model that activates 49 billion parameters, and its announcement page says 3,800 GPUs, not the 4,000 three outlets ran.
Read MoreGhost raised $11 million led by Andreessen Horowitz for Core, a $3,499 computer that runs AI agents at home. Its own privacy policy describes a Ghost gateway that forwards web search and phone call requests to third-party providers.
Read MoreReflection AI launched Beam on 5 October, a 501 billion parameter open-weight model, and the company’s own benchmark table puts
Read MoreORCA-bench puts five frontier coding agents in a production-fidelity oncall setting across 1,079 root cause analysis tasks. The best accuracy is 25.3% on realistic reports, and GLM-5 invents a root cause in 66.1% of the hardest ones.
Read MoreOpenAI pulled GPT-6.1 Astra the day before DevDay. The report published on the day of the event says the release was delayed indefinitely, not cancelled, and the four accounts disagree on when it was due.
Read MoreModulate raised $25 million on 28 September and said its transcription model recently topped Hugging Face Open ASR Leaderboard. The leaderboard version published three days earlier put it third.
Read MoreApple’s new speech paper reports three headline improvements over a bilingual baseline, and its own Table 1 shows they don’t
Read MoreOpenAI’s new Dots agents aren’t reaching its paying Pro subscribers in Europe. In a statement sent to press, the company
Read MoreServiceNow CoreAI says its AutoSynthData pipeline raised a 26B Gemma model from 18.77% to 27.18% on EnterpriseOps-Gym ITSM tasks. That is about 1.3 points below the best frontier score in ServiceNow own benchmark paper.
Read MoreFlorida’s motion to enjoin OpenAI runs 49 pages. It rests on four rogue agent incidents, not the three that Monday’s
Read MoreOpenAI gave every Pro subscriber a free Dot at DevDay, then in a separate announcement cut the Codex and Work allowance those agents run on from 20x Plus to 10x, effective 30 October.
Read More