ByteDance and Alibaba will shut down user-created AI companion features in their Doubao and Qwen apps respectively starting July 15, as China's new interim rules for anthropomorphic AI interactive ser…
A hyper-optimized, zero-dependency C/CUDA inference engine for the Qwen 3.6 35B model on RTX 5090 Blackwell GPUs achieves 13.4k tokens/sec prefill throughput at 2,048 context depth and 270+ tokens/sec…
Open-weight models have closed the quality gap with frontier systems, making the model itself the least defensible part of an AI business. Founders now build moats through workflows, context, and inte…
An engineer at a company built an internal benchmark to evaluate coding LLMs for production use, testing ten models on real-world tasks like Python services and JavaScript debugging. The results showe…
A solo developer running brevio.news discovered that a hosted LLM API silently broke the product's weekly brief feature for four weeks after the provider retired a model ID. The developer fixed the is…
A user tested the free local AI model Qwen 3 8B on a 16GB Apple Silicon Mac to reduce API costs, finding it effective for some tasks like summarization and writing but falling short on others, offerin…
Mesh LLM, a new open-source tool, launched yesterday and hit the Hacker News front page, enabling users to pool GPUs across machines and run AI models up to 235B parameters via an OpenAI-compatible AP…
A developer reported that running GPT-4o for a side project led to a $400 monthly AI bill, prompting a shift to Chinese models like DeepSeek V4 Flash and Qwen3-32B, which claim up to 40x cost savings.…
A human blogger built an AI slop machine despite writing every word of their blog manually. American and Chinese AI labs release increasingly strong open-weights large language models that can be self…
NovapAI provides a practical guide for integrating open-weight large language models like Llama, Mistral, Falcon, and Qwen into applications via a standardized API layer. The tutorial covers the HTTP …
A developer outlines best practices for debugging AI API failures in multi-model applications, emphasizing the need for failure classification, request logs, workflow-specific monitoring, and quality …
Thinking Machines Lab, led by Mira Murati, published a report arguing for human-centered AI built on customizable model weights. The lab proposes four technical directions to make AI distributed and s…
A developer fixed three bugs in the qMLX serving stack to make the Qwen3.5-122B model usable on a Mac Studio, reducing prefill latency from minutes to seconds for long-context conversations. The fixes…
A backend engineer's field notes from 2026 reveal a 350× price spread between the cheapest and most expensive AI APIs, with Qwen3-8B at $0.01/M output tokens and DeepSeek V4 Pro at $3.50/M. The engine…
A developer argues that C# is well-positioned for AI infrastructure, citing TensorSharp, a native C# inference engine that outperforms stable-diffusion.cpp in benchmark tests. The article draws parall…
An engineer running multi-region AI workloads reports that Chinese LLM APIs from DeepSeek, Qwen, GLM, and Kimi offer dramatically lower costs than US providers like OpenAI, Anthropic, and Google, with…
ARCQuant, a new framework for Large Language Model inference, uses the NVFP4 numerical format to achieve up to 3x speedup on GPUs while maintaining accuracy comparable to full-precision baselines. The…
Voicebox, a free and open-source AI voice studio, launches as a local-first alternative to ElevenLabs and WisprFlow, enabling voice cloning, speech generation in 23 languages, and dictation with full …
A Hacker News user asks the community to share specific tasks from the past month where open-weight models like GLM, DeepSeek, Kimi, or Qwen failed but frontier models like Opus, Fable, or GPT succeed…
Researchers have developed a new approach to overcome the quadratic cost of causal self-attention in long-context transformers, using state update design and structural interventions like sink tokens …