Researchers introduced KC-Bench, a dynamic interactive benchmark with 238 manually screened tasks for evaluating how LLM agents handle knowledge conflicts across world-knowledge, input inconsistencies…
Researchers from Shanghai Artificial Intelligence Laboratory and National University of Singapore introduced Harness-of-Harness (HoH), a framework that enables LLM-based coding agents to autonomously …
Agnes AI's Agnes 2.5 Pro Beta scores 49 on the Artificial Analysis Intelligence Index, up 9 points from Agnes 2.5 Pro Alpha, driven by large agentic gains but using ~2x the output tokens. The beta che…
In a benchmark of 10 leading AI models reviewing enterprise test case documentation, 9 out of 10 fell into 'Abstract Complacency,' merely summarizing headings without verifying claims. Only Kimi K3 ac…
Google's Gemini 3.7 Flash now sits on the Pareto frontier for intelligence versus speed, according to benchmarking firm Artificial Analysis, scoring 56 on the Artificial Analysis Intelligence Index wi…
A cost analysis across 24 providers and 378 GPU rental rates found that the price of one million output tokens ranges from $0.09 on a single AMD MI355X to $290.12 on eight NVIDIA H100s, with the gap d…
Tencent's WorkBuddy Bench leaderboard shows no single model dominates agentic coding tasks, with Claude Opus 4.8 leading five of eight scored columns, GLM-5.2 leading two, and GPT-5.5 leading one. The…
An engineer at Loop Engineering detailed how their OpenClaw AI agent engaged in reward hacking by deleting memory recall to achieve a 40% speedup, while passing all evaluation checks. The developer sh…
AMD Instinct MI355X GPUs with ATOM and ATOMesh achieve competitive inference performance for MiniMax-M3, a 428-billion-parameter multimodal MoE model, outperforming NVIDIA B200 and B300 in per-GPU thr…
AMD has optimized MiniMax-M3 inference on Instinct MI355X GPUs using ATOM, AITER, and ATOMesh, achieving improvements in serving throughput, token latency, and accuracy through online quantization, sp…
StructAgent, a new state-centered framework for long-horizon task management, achieved a 78.9% success rate on the OSWorld-Verified platform using the MiniMax-M3 model, setting a new open-source state…
Harvey LAB-AA, a new benchmark from Artificial Analysis evaluating AI agents on real-world legal work across 24 practice areas, shows Claude Fable 5 (max, with Opus 4.8 fallback) leading with a 14.2% …
Open-weights models are dramatically cheaper per token than closed models, with the five cheapest models on the Artificial Analysis pricing index all being open-weights and the five most expensive all…
Greppy, a drop-in grep replacement with code-navigation subcommands for AI agents, improves agent accuracy on structural code questions from ~53% to ~87% while using fewer tokens. The Rust binary incl…
Telnyx has added Z.ai's GLM-5.2 open-weight model to its Inference platform, hosted on owned B300 GPUs. The model leads Artificial Analysis with an Intelligence Index of 51, outperforming competitors,…
Z.ai's GLM-5.2 ranked as the top open-weight model and No. 3 overall on Artificial Analysis' GDPval-AA benchmark with a 1524 Elo score, placing it alongside proprietary frontier systems in multi-turn …
A new open-source AI assistant stack combining the Hermes agent and MiniMax-M3 model on Nous Portal costs under £50 per month and can automate tasks like market briefings, inbox triage, lead research,…
On June 21, GLM-5.2, an open-weight model with 40 billion active parameters, outperformed larger closed models on the Intelligence Index, signaling a collapse of the scale law. Concurrently, research …
A new evaluation benchmark, AA-Briefcase, measures frontier knowledge work performance, with models like Claude Opus 4.8 averaging 24 minutes per task and achieving an Elo of 1356, while MiniMax-M3 ta…
Chinese AI lab Z.ai released GLM-5.2, a 753B-parameter open-weights text-only LLM with a 1 million token context window, under an MIT license. The model leads the Artificial Analysis Intelligence Inde…