A developer built Aura, a desktop app that gives AI agents access to business data by replacing names, emails, phone numbers and account numbers with stable tokens before anything reaches a cloud mode…
A developer built archdiagram-bench, a benchmark that tests whether large language models can produce usable high-level architecture diagrams from a real code repository, using draw.io XML output and …
LlamaIndex launched OpenDocRouter, an API that parses documents into markdown using 10 open-source and frontier models, each run through a versioned recipe of prompts, processing, and settings. The se…
A developer built Cross-View Conservation, a Kaggle benchmark that tests whether language models can keep two representations of the same answer consistent as workloads grow, using deterministically g…
A developer built a Kaggle benchmark testing 11 frontier vision-language models on zero-shot OCR of ISO metric tire size codes from 51 curved tire sidewall images across 22 dimensions. Google Gemini 3…
A developer benchmarked 10 AI models as judges on Kaggle, planting errors in their own prior answers to test self-preference, and found that only two small OpenAI models graded their own mistakes more…
Google released Android Bench 2.0, adding long-horizon tasks (LHTs), agentic evaluation, and continuous scoring to its benchmark for AI models and agents on Android development tasks. The update repla…
A researcher tested four coding agents — Gemini 3.8 Flash, GPT Astra 6, Opus 5 and Fable 5 — each given a $100 budget to autonomously build an open-source PDF editor, and found bugs within a few click…
A five-model test published in September 2026 found that after one pushback, three of five AI models — DeepSeek V4 Pro, Claude Opus 5.5 and GPT-6.1 Sol — ultimately endorsed a user's stated career dec…
A developer compared how many requests a $10 budget buys across three LLM APIs using a fixed 4,000-input/1,000-output token workload, finding GPT-6 Luna yields roughly 11,111 requests, Gemini 3.8 Flas…
TypeSafe's Jev, a model that returns probability distributions over allowed answers rather than generating text, matched frontier models on claim verification at roughly 1/500th the cost of GPT-6 Astr…
A developer released mini-web-agent, an open-source Python project that implements a browser-controlling web agent in 396 lines of code, published on GitHub and installable via PyPI as mini-web-agent.…
A developer built CI Watchdog, an open-source GitHub App and SaaS platform that autonomously cancels redundant GitHub Actions runs and triages failed builds using open-weight AI models such as Gemma a…
A developer released Shlexball, a 1.5B-parameter local LLM that translates natural-language file descriptions into bash commands to locate files on Apple Silicon Macs, claiming it outperforms Gemini 3…
A developer's comparison of generative AI and agentic AI argues that generative AI remains the better default for most tasks a team ships this quarter, with agentic architectures justified only when w…
A developer's comparison found that generating complicated Rust functions on a local LLM took 30 to 60 minutes of 100% CPU and 100% GPU usage at an estimated electricity cost of $0.02 to $0.04, while …
GitHub deprecated four models across all GitHub Copilot experiences on October 2, 2026: Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7. GitHub recommends Gemini 3.8 Flash for …
Google announced Gemini 4 Argon, its newest frontier model with a 1-million-token context window built for advanced reasoning and cybersecurity defense, as the headline item in its September 2026 AI r…
A developer benchmarked four language models on 48 engineering-work logs to test whether they would falsely report "done" when a final check had failed or never ran. Without status definitions, Gemini…
A study of 503 profession-specific AGENTS.md profiles found that loading full scientific-agent profiles by default did not improve accuracy and cost 2.2 to 4.5 times more per successful call, accordin…