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
313

Are OpenAI's Mystery Aurora Models Image Generators? A Tester Thinks So

Three unannounced OpenAI model IDs, aurora-red-alpha, aurora-green-alpha and aurora-blue-alpha, were spotted in its API on October 18. A tester now says they look like image models and has posted generated images labeled red and green, and a tracking account agreed. OpenAI hasn't commented.

Aurora alpha models (unconfirmed)Read story
314

OpenCode Says a Security Incident Hit About 1.4% of Console Users

OpenCode says a security incident hit about 1.4% of Console users, all of whom had linked external providers. Notebookcheck reports its email warned saved provider keys may have been exposed. An Anthropic Claude Code team member says Claude keys believed affected were revoked.

OpenCode ConsoleRead story
315

Kimi's Website Code Lists a New 'k3d1-agent' Model: K3.1 or a 3D Model?

A script on Kimi's website lists an unannounced model ID, k3d1-agent, that is used only in a new make3d mode. One X user calls it Kimi K3.1, but the code doesn't confirm that, and Moonshot hasn't said anything.

Kimi k3d1-agent (unconfirmed)Read story
316

Elorian Launches Glimpse, a Visual Reasoning Model at $0.60/$3 per Million Tokens

Palo Alto startup Elorian has released its first model, Glimpse, which reasons over images and video by going back to the visual evidence as it thinks. It costs $0.60 per million input tokens and $3.00 per million output tokens, and Elorian says it performs at the level of GPT-6 Astra and Claude Opus 5.5 on multimodal benchmarks, sometimes better.

Elorian GlimpseRead story
317

DeepSeek V4.1 Flash 1017 Released With Open Weights, Same deepseek-flash API

DeepSeek released V4.1 Flash 1017 on October 17 with open weights. API users get it under the existing name deepseek-flash at unchanged prices, and DeepSeek says it scores higher while finishing tasks faster with shorter outputs.

DeepSeek V4.1 Flash 1017Read story
318

DeepSeek Search and Fetch API: Pricing, Free in Claude Code and Codex Until Nov 1

DeepSeek launched Search and Fetch APIs this week at $0.75 per 1,000 successful requests. Calls made automatically in Claude Code, Codex and DeepSeek Harness stay free until November 1.

DeepSeek Search API, DeepSeek Fetch APIRead story
319

Cline Desktop now runs cloud tasks in parallel, each in its own sandbox

Cline said Cline Desktop now runs in the cloud. You can run multiple tasks in parallel, each in its own secure cloud sandbox, and close your laptop while Cline keeps working.

Cline Desktop(云端并行任务)Read story
320

OpenCode adds Beam with a 1M context, free for a limited time

OpenCode now lists Beam Free on Console under the ID beam-free, with a 1M context and a limited-time free tier. The docs say the provider does not train on your data and keeps it for 30 days for safety; some regions are blocked.

Beam Free(OpenCode 免费档 beam-free)Read story
321

Cursor cloud agents can build shareable canvases; Environments API hits beta

Cursor cloud agents can now build interactive canvases such as dashboards or reports and share them by link. The changelog also lists a beta Environments API, up to 300 repositories per environment, and 5x faster loads for large ones.

Cursor Cloud Agents / Canvases / Environments APIRead story
322

OpenAI says Codex and Work now give more effective usage

On Day 12 of his update thread, OpenAI's Tibo said Codex and Work now deliver more effective usage by reusing context across subagents, keeping context cached longer, and improving request routing to raise cache hits.

Codex / ChatGPT WorkRead story
323

China proposes export controls on large models, AI agents and compute clusters, but says API services and already open-sourced tech aren't covered

A draft list from China's commerce and science ministries would add 23 AI technologies to its export control catalog. Officials say offering models through APIs, apps or SaaS isn't a technology export, and technology already open-sourced is excluded.

China AI export control catalogue (draft)Read story
324

VS Code 1.143 Makes the Copilot Harness the Default for New Sessions

VS Code 1.142 is gradually rolling out the Copilot harness as the default for new chat sessions, restores a read-only Ask agent, and splits the model picker into Copilot and BYOK tabs with per-model thinking level and context window settings.

VS Code 1.142–1.143 (GitHub Copilot)Read story