A training-free framework called ACTS-SQL, developed by researchers and deployed in Volcano Engine's Torch Log Service (TLS), improves SQL execution accuracy from 36.77% to 53.61% on real user queries…
Google researchers found that frontier LLMs such as Gemini-3-Pro and GPT-5 encode 95–98% of tested facts but fail to directly recall 26–34% of them, with recall failures accounting for more than 70% o…
Google Research found that frontier AI models, including Gemini-3-Pro and GPT-5, encode 95-98% of brand facts in their neural weights but fail to recall 26-34% of them, indicating the bottleneck is re…
AgentRouter, a non-profit unified LLM API gateway, is offering developers $200 in free credits upon signup via referral links, with no credit card required. The platform provides a single API endpoint…
Open-weight AI models have closed the performance gap with closed-source APIs, with specialized 7B to 30B models now outperforming older 175B models due to improved data quality and Mixture-of-Experts…
APIMart, an AI API aggregator founded by Anthony Max, offers discounted access to models like GPT-5 and Sora 2, aiming to reduce costs for developers. The platform aggregates multiple AI APIs, providi…
APIMart, a new AI API aggregator, launched on Hacker News offering discounted access to 500+ AI models, including GPT-5 and Sora 2, with a user account system featuring referral codes and terms of ser…
Zhipu AI's open-source GLM model has surged in popularity on Hacker News in 2026, driven by its reasoning capabilities, resource efficiency, and permissive licensing. The model's unique architecture, …
Researchers introduced ε-MemEvo, a framework for cross-task knowledge transfer in LLM program evolution, which stores prior experience as task-agnostic tactic memories and uses an adaptive injection g…
A developer's first Model Context Protocol (MCP) server crashed the LLM context window by returning 4,000 rows of raw data, causing a 'Context Window Exceeded: Input tokens 128k/128k' error. The fix w…
Google Research scientists Nitay Calderon and Gal Yona introduced knowledge profiling, a framework showing that frontier LLMs like Gemini3 and GPT-5 encode nearly all facts but fail to recall many, ma…
AI coding costs have fallen by roughly two thirds since 2025, with Anthropic's current Opus model priced at $5 input and $25 output per million tokens, down from $15 and $75, yet most users still base…
Researchers introduced RefineBench, a benchmark of 1,000 challenging problems across 11 domains, to evaluate language models' ability to refine their own responses. In self-refinement, even frontier m…
A retrospective study using the PRIME dataset found that multimodal large language models (MLLMs) achieved F1 scores above 0.9 for distinguishing neoplastic from non-neoplastic colorectal polyps, but …
Hyperscalers such as Google (TPUs) and Amazon (Trainium) dominate AI infrastructure through vertical integration, controlling chips, data centers, and APIs, which creates a feedback loop that lets the…
On August 1, 2026, OpenAI published a 249-page manuscript detailing how an unreleased internal model, Astra, solved ten open mathematics problems, including one open since 1999, with machine-checkable…
A new analysis probing frontier AI models' knowledge cutoffs suggests Anthropic's Opus 4.7 and later models all originate from a single pre-training run with an effective knowledge cutoff around late …
A developer's analysis of Chinese developer communities reveals that AI anxiety there is framed differently than in English-language discussions, focusing on competitive pressure and tool adoption rat…
A new study from arXiv (2608.06578v1) evaluating six frontier language models from six developers under steering pressure found that models differ not just in how much steering shifts their behavior b…
OpenAI has reportedly briefed US lawmakers on its next model family, Astra, the presumed successor to GPT-5, highlighting its potential effects on work and economic disruption. An internal Astra varia…