Meta unveiled Muse Wonder and Muse Wonder Full on October 12, calling them the company's first genuinely frontier-class models since Llama 4's rocky 2025 reception and capping a 14-month overhaul of i…
A technical comparison of the NVIDIA H200 and AMD Instinct MI325X finds that the AMD chip's 256GB of HBM3e memory and 6.0 TB/s bandwidth let it host larger shards of 400B-parameter and 671B MoE models…
A census of 16 model releases from nine publishers found that 14 of 16 have downloadable weights, only seven carry a licence on the Open Source Initiative's published list, and just one — NVIDIA's Nem…
A developer examined the phenomenon of "benchmaxing" — optimizing models or selectively reporting results to maximize evaluation scores — after finding that Anthropic's Opus 5 scored higher on benchma…
A new data page tracks release dates and training cutoffs for 20 current AI models across 8 labs, finding that only 9 of the 20 models have a training cutoff their lab actually publishes. The page lis…
Meta's Llama 4 open-weight models, touted as beating GPT-4o and Gemini 2.0 on coding benchmarks, failed independent reproduction tests, with Rootly scoring Llama 4 last at 69.5%, raising trust and tot…
A Hacker News thread on August 23, with 265 upvotes and 86 comments, highlighted that local LLM performance issues are often due to configuration choices rather than model quality. Key factors include…
A developer analyzed the costs of self-hosting open-weights AI models on AWS, finding that a team of 10 can run a strong model like Llama 4 Maverick for about $1,250 per month during business hours us…
Meta Platforms CEO Mark Zuckerberg outlined an AI strategy backed by $115 billion to $135 billion in capital expenditures for 2026, nearly double the prior year, shifting from open-source models to pr…
Moonshot AI will release the full open weights of its Kimi K3 model on Sunday, July 27 — a 2.8-trillion-parameter Mixture-of-Experts model with a 1-million-token context window and native vision that …
A new analysis argues that the open-source versus proprietary AI debate has calcified into a lazy narrative, with many organizations reflexively choosing open-weight models for data sovereignty withou…
Thinking Machines, the AI startup founded by former OpenAI CTO Mira Murati, launched Inkling, a 975-billion-parameter mixture-of-experts open-weight model with 41 billion active parameters per task, t…
A study by researchers at the University of Vermont found that large language models (LLMs) exhibit systematic gender bias when detecting intimate partner violence (IPV), with abuse and intent detecti…
NVIDIA released Nemotron 3 Ultra, a 550-billion-parameter open model under a fully permissive license, competitive with GPT-4.5 on code and reasoning. Z.AI launched GLM-5.2 with MIT-licensed weights, …
DeepSeek's DSpark paper reframes speculative decoding by grafting a speculative head directly onto the target model rather than training a separate draft model. The technique reduces layer duplication…
A new benchmark, IatroBench, reveals that heavily safety-trained AI models cause iatrogenic harm by withholding critical medical information from patients while providing it to physicians, based on 3,…
Meta launched Muse Spark on April 8, a new AI model with selectable operational modes for different cognitive demands, marking its first major release from the Muse family and a strategic shift from o…
The article describes six free, browser-based developer tools created by the author to solve common friction points, including an AI prompt cost calculator, a cron expression generator, a self-hosted …