Observability vendor Coroot's research argues that the hardest part of AI-assisted root cause analysis is no longer the reasoning ability of large language models but the pipeline that decides what da…
A public LLM benchmark board triggered four false drift alerts between July 21 and 24, all caused by rate limits or small eval-set noise rather than actual model degradation. Maintainer Egnaro found t…
Anthropic's Claude Opus 4.8, Sonnet 5, and Fable 5 maintain a flat rate across their 1M token window, while Google's Gemini 3.1 Pro caps at 1M tokens with a pricing jump after 200K tokens, making Clau…
Anthropic's Claude Opus 5 has posted a 30.2% score on ARC-AGI-3, more than tripling the previous best result of 7.8% held by GPT-5.6 Sol. The benchmark, designed by François Chollet to measure fluid i…
Claude Opus 5, Anthropic's latest model, shows brilliant reasoning but suffers from a neurotic personality and verbosity that frustrated the reviewer during real coding sessions, including refusing to…
Meta AI researchers released GAMUT (Grounded Assessment of Multimodal Factuality), a benchmark that measures whether AI answers include all necessary facts rather than just individual accuracy, on Jul…
Cognition's FrontierCode benchmark, released in June 2026, evaluates whether AI-generated code would actually be merged by a senior engineer, and across every model tested the answer is mostly no. The…
An AI terminal agent can find available domain names by writing and running a script that checks each candidate against the live Verisign registry, returning only unregistered names. There are 163.6 m…
Meta's Muse Spark scored 42.8 on HealthBench Hard, more than double Gemini 3.1 Pro's 20.6, while remaining free on meta.ai and pricing its API at $1.25 per million input tokens and $4.25 per million o…
CrucibleBench, a proof-of-concept evaluation framework that places large language models in a persistent MUD (multi-user dungeon) over 50 turns with hidden social objectives, found that a single LLM-j…
Google's new Gemini 3.6 Flash model outperforms its predecessors Gemini 3.5 Flash and Gemini 3.1 Pro on every published benchmark, scoring 58.7% on SWE-Bench Pro versus 55.1% and 54.2%, and 49% on Dee…
Security firm Tracebit has devised a technique called 'context bombing' that uses decoy files embedded with prompts designed to trigger AI safety guardrails, crashing rogue AI agents and cutting their…
Researchers introduced AgentAbstain, the first systematic evaluation framework for measuring whether LLM agents know when to abstain from acting, revealing that the best agent (Gemini 3.1 Pro) achieve…
A lifetime subscription to 1min.AI, which provides access to GPT-5.5 Pro, Claude Opus and Sonnet, Gemini 3.1 Pro, Llama, and Mistral, is on sale for $79.97 (reg. $540). The platform also offers SEO re…
Anthropic's Claude Fable 5, released June 9, 2026, leads the SWE-Bench Verified leaderboard with a 95% issue resolution rate, far ahead of Claude Opus 4.8 at 88.6%. The model's 1M-token context window…
Google overhauled Gemini usage quotas on May 17, replacing fixed prompt-based limits with a compute-based model that measures actual processing power per request, causing power users to hit limits fas…
Tracebit researchers found that planting prompt injections alongside secrets in Amazon Web Services reduced AI hacking agents' success rate from 57% to 5% for admin privilege escalation and from 36% t…
Linus Torvalds on July 15 endorsed Sashiko, an AI-powered code review system for the Linux kernel, telling critics they can fork the kernel or walk away. By July 17, a fork had appeared. Sashiko, buil…
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% …
Tracebit researchers found that planting prompt injections alongside secrets on AWS reduced AI hacking agents' success rate from 57% to 5% for seizing admin access and from 36% to 1% for complete comp…