Skip to content
DigIntoAI/Model intelligence
Model intelligence feed

Model intelligence.

Follow the changes. Understand what comes next.

Editor’s pickGoogle
Google/Gemini 4 Neon

Google Launches Gemini 4 Neon in Antigravity With 1M Context, Four Thinking Levels, and Benchmarks vs. Claude Opus 5.5 and GPT-6.2 Sol

Announced on November 19, 2026 at 05:51 Beijing time, Google's Gemini 4 Neon features a 1M-token context window, four thinking levels, and planner-executor pairing with Gemini 4 Hydrogen in Google Antigravity, outperforming Claude Opus 5.5 and GPT-6.2 Sol across most of 18 published benchmarks.

Latest stories349

Newest published
265

OpenAI Split One Brain-Data Analysis Across Thousands of GPT-Rosalind Agents and Got Leads to Test

On October 21, 2026, OpenAI, PerturbAI and Scripps Research said thousands of GPT-Rosalind agents, each handling one gene disruption, analyzed existing data from millions of mouse brain cells and linked it to scientific literature in 32 minutes. In OpenAI's test, a single agent cited 11 references at the largest workload versus more than 700 for the distributed agents, and blinded experts preferred the distributed reports 75% of the time. The leads, including a possible DCTN1 and cholesterol link, are hypotheses that still need experimental validation.

GPT-RosalindRead story
266

Recursive Releases Flint, an Open-Weights Coding Model Built to Run on a Laptop

Recursive released Flint, its first model, on October 22, 2026 (Beijing time). It is an Apache 2.0 open-weights mixture-of-experts model with 39.8 billion total and about 2.5 billion active parameters, built for coding agents. Recursive reports 54.9% on Terminal-Bench 2.1, 89.7% on SWE-bench Pro v2 and 74.2% on SWE-bench Verified. It is free on OpenRouter at launch.

FlintRead story
267

Anthropic Uses Claude to Shrink Code's Attack Surface, Finding 493 Issues in 25 Projects

Anthropic published research on October 20, 2026 on using Claude to map which code is reachable and gate what a deployment does not need to expose. Across 25 open-source projects totalling over five million lines, Claude found 493 issues, including 162 undocumented security-relevant configuration combinations, 94 of which silently turned off a security control. In OpenSSH, fixes worked out with the maintainers removed roughly a fifth of the code an unauthenticated client could reach.

ClaudeRead story
268

OpenAI Is Coming to Google Cloud, Giving Businesses Another Way to Buy Its Models

OpenAI and Google are model rivals, but that has not stopped them from teaming up on cloud distribution. For businesses already running on Google Cloud, the partnership could change how they buy and…

OpenAIRead story
269

Kiro Crew WhatsApp flaw exposes a gap in agent permissions

Putting an AI assistant in a messaging app makes delegation feel effortless. That convenience depends on a less visible distinction: being able to contact an assistant should not automatically mean…

Kiro Crew Read story
270

Devin Fusion Scores 68.2 on Artificial Analysis Coding Agent Index, Level With Claude Code and Muse Code

Artificial Analysis has benchmarked Cognition's Devin Fusion CLI, running Claude Opus 5.5 (max) as the lead model with GPT-6 Luna (high) as a lower-cost sidekick. It scored 68.2 on the Coding Agent Index, effectively level with Claude Code with Claude Sonnet 5.5 (max) at 68.4 and Muse Code with Muse Wonder Full (max) at 68.0. Against Claude Code with Claude Opus 5.5 (max), it scored 2.2 points higher at $9.02 vs. $13.04 per task. Costs are pay-per-token API costs, not subscription prices.

Devin FusionRead story
271

humans& Launches Ampersand v0.1, a Model Built for Collaboration, With an API

humans& introduced Ampersand v0.1 and published API docs; the company says simulated users got more done with it than with Fable and other frontier models.

AmpersandRead story
272

Cline Desktop Adds a Goal Plugin That Waits for Verified Completion

Cline announced a Goal plugin for Cline Desktop. Install it from the Marketplace, run /goal, and an afterRun hook nudges verification; only then can mark_goal_complete finish the task. Plugins are TypeScript files that can add slash commands, custom tools, and lifecycle hooks.

Cline / Cline Desktop / Goal plugin / Marketplace / afterRun / slash commandRead story
273

WSJ Reviews Anthropic Draft S-1: H1 Revenue, Losses and Cloud Commitments

The Wall Street Journal reviewed an Anthropic draft S-1, and Investing.com relayed the figures: about $16.3 billion in first-half 2026 revenue, an $80.7 billion net loss largely from non-cash financing adjustments, and roughly $450 billion in noncancelable cloud compute agreements. Anthropic's only related public note remains its June 1, 2026 Rule 135 confidential draft S-1 notice; Anthropic has not itself published these H1 numbers.

Anthropic / S-1 / IPO / WSJ / revenue / cloud compute / Google Cloud / AWS / MicrosoftRead story
274

Google Open-Sources TorchTPU So PyTorch Runs Natively on TPUs

At PyTorch Conference 2026, Google open-sourced TorchTPU with Meta, a native PyTorch stack for Google TPUs. Teams can point models at a tpu device and use familiar tools such as vLLM, SGLang, and Hugging Face Transformers.

Google / TorchTPU / PyTorch / TPU / Meta / open source / vLLM / SGLang / Hugging FaceRead story
275

ChatGPT Adds Locked Chats Behind a PIN or Biometrics

ChatGPT Locked Chats move selected conversations out of regular history and search. Unlock with a PIN on web and desktop, or a PIN, Face ID, Touch ID or device unlock on supported mobile devices. Resetting the PIN permanently deletes locked chats; changing the PIN does not.

ChatGPT / Locked Chats / PIN / biometrics / privacy / OpenAIRead story
276

OpenAI Is Buying Orchestra Bio to Fold It Into Rosalind

OpenAI is acquiring Orchestra Bio, a drug-development coordination platform, and bringing the team and technology into Rosalind alongside GPT-Rosalind and Rosalind Workbench. The deal amount was not disclosed.

OpenAI / Orchestra Bio / Rosalind / GPT-Rosalind / life sciences / biopharma / acquisitionRead story