Model intelligence.
Follow the changes. Understand what comes next.
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 publishedOpenAI Introduces Computer-Use Agents, Upgraded Clinical Search, and 20+ Healthcare Plugins for ChatGPT
On November 16, 2026, OpenAI announced new capabilities across ChatGPT for Healthcare and ChatGPT for Clinicians (free for verified U.S. clinicians), introducing computer-use agents that navigate existing healthcare software, improved trusted clinical search, and more than 20 new healthcare plugins. OpenAI also published workflow evaluations for GPT-6 Astra and GPT-6 Luna alongside HIPAA/BAA governance and human-in-the-loop review controls.
Qwen4-Max Official API ID (qwen4-max) Launches: 2.4T Architecture, Regional USD Pricing & OpenRouter / OpenCode Go Access
On November 16, 2026, Qwen introduced the official API model ID qwen4-max for its flagship Qwen4-Max model, distinct from the limited-time qwen4-max-preview launched on November 11. Built on a 2.4T sparse MoE backbone (71B active per token) plus 328B N-gram Embeddings with a 1M context window, qwen4-max is priced at $1.50 input / $4.50 output / $0.075 cache read per 1M tokens in Singapore ($1.413 / $4.24 / $0.071 in Beijing and Global regions) and is now live on OpenRouter and OpenCode Go.
Vercel Token Pass Raises Monthly Allowance from $80 to $150 at $30/Month and Adds Kimi K3.1 Pro
Across November 12 and November 13, 2026, Vercel updated its AI Gateway add-on Vercel Token Pass—increasing the monthly token allowance from $80 to $150 per team at the same $30/month price (resetting each billing cycle with no rollover) and adding Kimi K3.1 Pro (moonshotai/kimi-k3.1-pro) to its curated open-weight model catalog with zero data retention.
Cloudflare Workers AI Adds Multimodal Embedding Model EmbeddingGemma 2 (@cf/google/embeddinggemma-2)
On November 12, 2026, Cloudflare announced that Google DeepMind's open multimodal embedding model EmbeddingGemma 2 (@cf/google/embeddinggemma-2) is available on Workers AI. Mapping text, code, images, video, and audio into a shared 768-dimensional vector space with an 8,192-token context window and flexible 128/256/512 dimensions, it is priced at $0.0017 (or 154 neurons) per million input tokens.
Parallel Decisions Fast Launches on Vercel AI Gateway, Free Through November 23, 2026
On November 13, 2026, Vercel added parallel/parallel-decisions-fast to Vercel AI Gateway with free usage through November 23, 2026. Accessible with an AI Gateway API key and no separate Parallel account, the 32K-context model searches the live web and returns calibrated probabilities with cited evidence across Boolean, Choice, and Score questions.
Cohere Completes AISLE Acquisition, Welcomes cubic Team, and Launches Cohere Cyber
Following its October 15, 2026 letter of intent, Cohere announced on November 13, 2026 the completion of its acquisition of autonomous vulnerability discovery and remediation company AISLE, alongside welcoming AI code-review platform cubic and establishing Cohere Cyber. Jaya Baloo joins as Cohere's first CISO, while Cohere plans long-term expansion in Prague for sovereign AI security.
Grok 4.7 Launches with Experimental Support on Kiro in AWS GovCloud (US)
On November 13, 2026, AWS announced that Grok 4.7 is now available with experimental support in the Kiro IDE and CLI across AWS GovCloud (US) Regions. Described by AWS as SpaceXSI's most capable model for coding and knowledge work, it offers a 500K context window billed at a 2.7x credit multiplier up to 200K tokens and 5.4x above 200K tokens, subject to administrator enablement.
NVIDIA NIM for LLM/VLM 2.1.6 Released: Five Models Enter Mainline Containers, Backend Upgraded to vLLM 0.32.0 & SGLang 0.5.22
NVIDIA has released NIM for LLM/VLM 2.1.6 (November 14, 2026). The update upgrades inference backends to vLLM 0.32.0 and SGLang 0.5.22, promotes deepseek-v4.1-flash, inkling-small, muse-glimmer, nemotron-3.5-super-vl-120b-a12b, and qwen3.8-27b to their first main releases, adds cached runtime manifests and optional startup warmup requests, and patches 99 unique security issues.
Runway Releases Open Weights for Praxis-1: World Action Model Architecture, Access Rules & Benchmarks
Following its September 2026 research introduction, Runway released open weights and inference/fine-tuning code for its Praxis-1 world action model under Apache License 2.0 on November 11, 2026. Designed to turn video pretraining into physical robot control rather than generating film clips, Praxis-1 spans single-arm DROID, bimanual YAM, and SO-101 variants. Here is a verified breakdown of its architecture specs, Hugging Face auto-gating download rules, 80 GB VRAM quickstart floor, and self-reported RoboLab-120 results.
n8n Labs Introduces Nod (Research Preview): Standalone Personal Desktop Agent Requirements, Licensing & Spending Controls
On November 10, 2026, n8n Labs introduced Nod, a standalone personal desktop agent released as a research preview (with GitHub tag nod-v0.7.6 published on November 14, 2026 Beijing time). Separate from core n8n and requiring no n8n account, Nod runs locally on macOS, Windows 11 preview, Linux x64 Flatpak preview, or a single-user self-hosted server. This report breaks down its platform requirements, bring-your-own-model rules, three separate spending controls, Apache 2.0 vs. GPL-3.0-or-later license boundaries, and self-reported LongMemEval-S benchmark context.
Perplexity Decisions API Adds pplx-decider-v1.3, flash-v1.3, and Text-Span Matching via locate
In its November 2026 API changelog, Perplexity introduced pplx-decider-v1.3 (labeled 27B) and pplx-decider-flash-v1.3 (labeled 4B) for the Decisions API, updating the pplx-decider-latest and pplx-decider-flash-latest aliases. Retaining input-only pricing of $0.02 and $0.01 per 1M input tokens with free output tokens, the v1.3 release adds the extractive locate question type for string states, while the local Decider for Mac app remains documented on flash-v1.2.
Third-Party Models Failing After Upgrading to Claude Code 2.1.323? How to Fix with CLAUDE_CODE_DISABLE_1M_CONTEXT=1
If you route Claude Code through OpenRouter, LiteLLM, custom AWS Bedrock endpoints, or third-party API gateways and started seeing every turn fail mid-session right after upgrading to v2.1.323…