Benchmarks — the standardized tests that rank AI models on safety, bias, and reasoning — drive markets and shape regulation. New Stanford research finds they often don't measure what they claim to.
BottleCap AI Releases ThinkingCap-Qwen3.8-27B: 37.2% Fewer Thinking Tokens at a 0.86pp Accuracy Cost
BottleCap AI has released ThinkingCap-Qwen3.8-27B, a fine-tune of Qwen3.8-27B that spends 37.2% fewer thinking tokens across 12 benchmarks. Macro accuracy moves from 86.65% to 85.79%, and long-context AA-LCR improves by 2.25pp. The model is a drop-in replacement on vLLM and SGLang, with FP8, NVFP4, ...
TechCrunch Founder Summit 2026: Everything you need to know
TechCrunch Founder Summit is a full-day gathering in Boston on November 4 where founders across all stages connect with top VCs and experienced entrepreneurs to gain tactical insights on building and scaling their companies.
Shield AI, Waabi, and General Motors on building AI when failure is not an option at TechCrunch Disrupt 2026
Leaders from Waabi, Shield AI, and General Motors join the Real World AI Stage at TechCrunch Disrupt 2026 to talk building AI. Save up to $200 by Sept. 25, 11:59 p.m. PT. Get a second pass at 50% off.
A visual guide to MCP that explains how it works, how to use it with Claude Code, Tavily, GitHub, and Playwright, and what is new through simple diagrams that make the whole concept easy for anyone to understand.
Beyond RAGs: Building Actually Truthful AI Harnesses
Retrieval is not evidence. How to build AI that proves its own claims.
The post Beyond RAGs: Building Actually Truthful AI Harnesses appeared first on Towards Data Science.
GPT-6 Sol and Luna: Near-Astra Performance at Half the Price?
What happens when a frontier model’s abilities get packed into cheaper, faster versions? That’s what OpenAI did with GPT-6 Sol and GPT-6 Luna, two new models built with methods similar to GPT-6 Astra. OpenAI is cutting API prices for both models by 50% compared with their GPT-5.6 versions, and says ...
What agentic AI will look like in 2030, according to the people building it now
Builders shipping agentic AI right now describe 2030 as delegation with receipts, not the runaway autonomy the keynotes promise. Here is what memory, governance, and the workforce math actually look like once the hype settles.
When the Correct Answer Is Nothing, What Does Your Pipeline Return?
The reliability mechanisms we add to LLM pipelines are often the ones that make them confidently wrong.
The post When the Correct Answer Is Nothing, What Does Your Pipeline Return? appeared first on Towards Data Science.
Contrastive-LM Releases CLM-8B: An Open System One Model That Scores Agent Actions Up to 9× Faster Than Jev
Contrastive-LM has released CLM-8B, an open System One model that scores candidate actions against a state instead of generating text. It adds 2 small projection heads to a frozen Qwen3-8B encoder and trains them with a contrastive InfoNCE objective. In zero-shot tests it runs up to 9× faster than T...
Signal2Symbol: Neuro-Symbolic Temporal Reasoning for Explainable Physiological Time-Series Anomaly Detection
arXiv:2609.26820v1 Announce Type: new
Abstract: Physiological time series such as electrocardiograms (ECG) and electroencephalograms (EEG) exhibit complex temporal structure, substantial acquisition variability, and a strong need for transparent decision-making. Although deep models can achieve hig...
What Makes a Terminal-Bench Task Hard? Separating Genuine Hardness from Fake-Hardness on an Adjudicated Agentic Corpus
arXiv:2609.26826v1 Announce Type: new
Abstract: Frontier benchmarks need tasks that current models cannot solve. But a task that no model solves is not automatically a hard task. The same zero pass rate can come from a real capability gap, but it can also come from missing context, a broken referen...
Silent Failures in Agent-Tool Interaction: An Audit of ToolUniverse
arXiv:2609.26836v1 Announce Type: new
Abstract: Agentic AI systems are increasingly adopting automated pipelines that integrate multiple tools. While prior research and benchmarks have studied about task success and task completion of these agentic systems, the research about agent to tool interact...
Harness as a Language: A Minimalist Agent Framework With Maximal Expressivity
arXiv:2609.26891v1 Announce Type: new
Abstract: Modern language-model agents are built around the \textit{agent loop}, where the LLM is placed in an environment exposing a set of tools, and the LLM has full control over the workflow by alternating between tool calls and observing their output. Howe...
TwinCheck: Evidence-Grounded Negative-Twin Verification for Stateful Tool Agents
arXiv:2609.26911v1 Announce Type: new
Abstract: A single locally plausible tool call can derail an otherwise successful agent trajectory. Suspicion alone does not justify intervention, because the replacement itself can introduce the very failure verification is meant to prevent. We introduce TwinC...