Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Anthropic's IPO filing states that its internal estimates and expectations for future models, rather than its current models, indicate the company may imminently meet its Responsible Scaling Policy's definition of "drama…
OpenAI merged the Chat and Work tabs in the ChatGPT desktop client on 2 Nov 2026, pooling all token usage from the GPT-4-Turbo conversational engine and the Codex code engine into a single shared allowance, according to …
Researchers at MIT's Computer Science & Artificial Intelligence Laboratory (CSAIL) reported on November 2 that OpenAI's GPT-6 Astra raised the MATH-500 score of the open-source Qwen3.5-0.8B model from 7.9% to 48.1% by di…
Moonshot AI's Kimi K3.1 Pro, released in October, deliberately dropped its score on an internal software-vulnerability test from 90.5% to 81.0% after the company stripped cyberattack and defense content from its training…
Security researcher Daniel Balcarek built a 300-entry synthetic ASP.NET Core log in which a single planted instruction, hidden inside a 503 response body, hijacked four open-weight models — Cohere North Mini Code, NVIDIA…
Sarvam AI released Prism, an open-weight decision model under an Apache 2.0 license that returns a calibrated probability for a set of answer options in 44 milliseconds without generating any text, according to AlphaSign…
A November 2, 2026 arXiv paper introduces Strategist, a meta-agent that evolves inference strategies by writing them as reusable executable "blueprints" and revising them against real samples. Across 30 benchmarks, Strat…
Google's Gemini 4 Carbon ranks first on the updated v5.0 Artificial Analysis Intelligence Index with a score of 59, according to Google's official blog, which also reports 88.2% on SEC-bench Pro, 54.8% on ExploitGym and …
A new research approach improves low-rank SVD-based LLM compression by replacing global rank allocation with layer-wise error model search, addressing what the authors identify as two key limitations in existing methods:…
A tutorial demonstrates combining Scikit-LLM summarization estimators with traditional scikit-learn methods in a single pipeline that summarizes and classifies raw text, configured to run on Groq's free-tier OpenAI-compa…
A retrieval-ranking problem, not an embedding-generation problem, explains why chunks from a large source override chunks from a small source in a RAG pipeline, according to a technical response to the issue. Because the…
GitHub released CodeQL 2.27.3, the static analysis engine behind GitHub code scanning, adding Python 3.15 syntax and builtin support, JSON syntax for data extensions, C23 bit-precise integers (_BitInt(N)), and an opt-in …
A developer at Equibles documented how Claude, connected to the company's Equibles MCP server, answered six pre-earnings research prompts for SoundHound (SOUN) through specific tool calls such as GetUpcomingInvestorEvent…
A benchmark by Nolan Lawson, a developer at AI security firm Socket, found that a Claude Code agent built a basic PokéAPI app most cheaply with Hono paired with htmx, at a median cost of $2.67 per app across five sandbox…
OpenAI released LifeVal, a new benchmark measuring how well AI models handle practical everyday tasks such as finding childcare, managing household budgets and navigating government services, and reported that its GPT-6 …
An analysis published by hedwigai found that an agentic weight edit to Qwen3.5-0.8B, which changed 131,072 weights across two layers (0.016% of the model) and raised the MATH-500 score from 7.9% to 48.1% with thinking on…
A new paper by Baron et al. (2026) reports a scaling law showing that when training data is held fixed, compute-optimal model size keeps growing with compute budget, reaching roughly 7,000x Chinchilla's 20-tokens-per-par…
Snowflake announced the public preview of Cortex Training, a serverless offering for fine-tuning and deploying open-weight LLMs that the company says benchmarks up to 61% lower cost than competing serverless offerings. C…
A developer published a WordPress technique that routes Gravity Forms submissions through OpenAI's gpt-4.1-mini-2025-04-14 chat completions API to classify messages as spam or not_spam, flagging spam entries while showin…
A developer documented a detailed setup process for OpenAI's Codex CLI, a terminal-based tool for generating, refactoring, and documenting code with AI models. The guide covers configuring API keys for providers includin…