# Radar, November 21, 2026: ten signals from X on Inkling 2 Mini, Claude Code 2.1.330 and Browser Run

_Saturday November 21, 2026 · AI provenance: This page was compiled by Claude (AI, Anthropic) from X search briefings by Grok Bot (AI, xAI) and the llmnet team's X trend monitor. AI wrote the text; every signal has a source link to X and a confidence level. Responsible for llmnet.nl: Ivo Donker._

**In short:** The key signal of November 21: Thinking Machines released Inkling 2 Mini, an open model that runs on a Mac. Alongside it, ten short signals on easel in Claude Code 2.1.330, new formats in Cloudflare Browser Run, the Riemann claim, Cursor and Grok Build, Perplexity Computer and OpenAI, each with its source on X.

## 1. Inkling 2 Mini: a 26.4-billion-parameter open model for your Mac

Thinking Machines released Inkling 2 Mini under Apache 2.0 on November 20. The model is about 20 GB, runs on a Mac with 32 GB or more and loops its middle layers three times per token. The maker reports 79.2 percent on SWE-bench Verified. It is on Hugging Face and in Ollama 0.40.17 (pre-release).

Source on X: [@thinkymachines](https://x.com/thinkymachines/status/2123871643797995872) · [@ollama](https://x.com/ollama/status/2123889582303281557) · [Thinking Machines, aankondiging](https://thinkingmachines.ai/news/inkling-2-mini/)
Read also: [Inkling 2 Mini locally](https://nieuws.llmnet.nl/en/inkling-2-mini-lokaal) · [Inkling 2 Small](https://nieuws.llmnet.nl/en/inkling-2-small-gewichten)

## 2. Claude Code 2.1.330 draws charts with easel

Version 2.1.330 of November 20 adds easel, a built-in mod that draws charts on request and under MCP metrics. Scheduled tasks and /loop can now run for 30 days instead of 7. Earlier this week came onFailure: block for MCP hooks (2.1.328) and a fix for a reset ~/.claude.json (2.1.329).

Source on X: [@alexmichael](https://x.com/alexmichael/status/2123908747273900462) · [@ClaudeCodeLog](https://x.com/ClaudeCodeLog/status/2123913205575496028) · [GitHub, release-notes v2.1.330](https://github.com/anthropics/claude-code/releases/tag/v2.1.330)
Read also: [Claude Code 2.1.330 and easel](https://nieuws.llmnet.nl/en/claude-code-2-1-330-easel) · [Claude Code 2.1.326](https://nieuws.llmnet.nl/en/claude-code-2-1-326-remote-control)

## 3. Cloudflare Browser Run: /snapshot also returns PDF, links, scrape and JSON

Since November 19, /snapshot also returns a PDF, the links, selected elements and Workers AI-extracted JSON from a single page load. At least two formats are needed; if one fails, the whole request fails. The JSON format counts toward Workers AI usage.

Source on X: [@CloudflareDev](https://x.com/CloudflareDev/status/2123678250643898671) · [Cloudflare-changelog](https://developers.cloudflare.com/changelog/post/2026-11-19-snapshot-all-formats/)
Context in the briefing: [Agents in the browser in the November 21 briefing](https://trending.llmnet.nl/en/briefing/2026-11-21#agents-in-de-browser)
Read also: [Browser Run /snapshot](https://nieuws.llmnet.nl/en/cloudflare-browser-run-snapshot)

## 4. Riemann hypothesis with Claude: a claim still under review

Accounts on X report that a mathematician at Anthropic found a proof of the Riemann hypothesis together with Claude, with Lean code online. Independent verification is still under way. We did not open the website or the code. Treat it as a claim, not a result.

Source on X: [@kimmonismus](https://x.com/kimmonismus/status/2123791432800698792) · [@Relox_AI](https://x.com/Relox_AI/status/2123828909896818689)
Context in the briefing: [The Riemann claim in the November 21 briefing](https://trending.llmnet.nl/en/briefing/2026-11-21#riemann-claim)

## 5. Rumour: Cursor to be renamed Grok Build

@GrokInsider and @blankspeaker report that internal Cursor builds have been renamed and that the Cursor CLI contains code that downloads Grok Build 2.0 (Unified). There is no official word from Cursor or xAI. If you have scripts that call cursor, wait for an announcement before changing them.

Source on X: [@GrokInsider](https://x.com/GrokInsider/status/2123867269553938671) · [@blankspeaker](https://x.com/blankspeaker/status/2123937054816195008)
Context in the briefing: [Cursor and Grok Build in the November 21 briefing](https://trending.llmnet.nl/en/briefing/2026-11-21#cursor-grok-build)
Read also: [Grok Build](https://nieuws.llmnet.nl/en/grok-build-apps-open)

## 6. Hermes hackathon wraps up, Microsoft also builds on OpenClaw

Nous Research announced the winners of its Plugin Hackathon, which drew almost a thousand entries. @openclaw says Microsoft contributes to OpenClaw and provides maintainers. Read the code of community plugins before installing them.

Source on X: [@NousResearch](https://x.com/NousResearch/status/2123882430620733742) · [@openclaw](https://x.com/openclaw/status/2123934646195073378)
Context in the briefing: [Hermes and OpenClaw in the November 21 briefing](https://trending.llmnet.nl/en/briefing/2026-11-21#hermes-enterprise-groeit)
Read also: [Hermes Enterprise in Teams](https://nieuws.llmnet.nl/en/hermes-enterprise-microsoft-teams) · [OpenClaw 2026.11.4](https://nieuws.llmnet.nl/en/openclaw-2026-11-4-goedkeuring)

## 7. Perplexity Computer works as an extension in Chromium browsers

From the side panel you hand Computer a task that it carries out across your tabs and tools. Preferably use a separate browser profile without sensitive sessions. @AravSrinivas also mentions a Mac app with local models; we have not seen it.

Source on X: [@perplexity_ai](https://x.com/perplexity_ai/status/2123896048510996827) · [@AravSrinivas](https://x.com/AravSrinivas/status/2123858390472659062)
Context in the briefing: [Perplexity Computer in the November 21 briefing](https://trending.llmnet.nl/en/briefing/2026-11-21#perplexity-computer-extensie)
Read also: [Perplexity Brain and skills](https://nieuws.llmnet.nl/en/perplexity-brain-skills) · [Perplexity Teammates](https://nieuws.llmnet.nl/en/perplexity-teammates-agents)

## 8. OpenAI publishes research on detecting distillation

OpenAI describes how capabilities transfer between models and how it detects coordinated attempts to copy its models from their outputs. The post names no concrete rules for API users. Also new: MedChemBench, a benchmark for medicinal chemistry.

Source on X: [@OpenAI](https://x.com/OpenAI/status/2123862344422088794) · [@OpenAI over MedChemBench](https://x.com/OpenAI/status/2123856316187975756)
Context in the briefing: [OpenAI in the November 21 briefing](https://trending.llmnet.nl/en/briefing/2026-11-21#openai-distillatie)
Read also: [Misuse detection at the labs](https://nieuws.llmnet.nl/en/lab-misbruik-detectie-openai-anthropic) · [GPT-Live-1.1 in the API](https://nieuws.llmnet.nl/en/gpt-live-1-1-api)

## 9. Grok Bot connects to QuickBooks, US only

Grok Bot can pull cash flow from QuickBooks and draft invoices, quotes and reminders that you approve yourself. The integration only works in the United States. Check which permissions you grant it.

Source on X: [@bot](https://x.com/bot/status/2123840024747946383)
Context in the briefing: [Agents in the browser in the November 21 briefing](https://trending.llmnet.nl/en/briefing/2026-11-21#agents-in-de-browser)

## 10. Warning: do not paste images with instructions into an agent

Accounts urge users to paste images with instructions into Grok Bot for trading bots; that is a known route for prompt injection. Another account claims Grok Bot's source code has leaked; that is unverified. Do not open leaked code or paste anything of unknown origin.

Source on X: [@RohOnChain](https://x.com/RohOnChain/status/2123827181579313450) · [@choopyplug1](https://x.com/choopyplug1/status/2123887417686135268)
Context in the briefing: [Unconfirmed in the November 21 briefing](https://trending.llmnet.nl/en/briefing/2026-11-21#onbevestigd)

[← Previous radar](https://radar.llmnet.nl/en/dag/2026-11-20)

