Radar, November 20, 2026: ten signals from X on Hermes in Teams, GPT-Live-1.1 and Perplexity Brain
In short: The key signal of November 20: Nous Research's Hermes Agent now works in Microsoft Teams, the Microsoft Store and Azure. Alongside it, ten short signals on GPT-Live-1.1 in the API, self-written skills in Perplexity Brain, Anthropic's first trading day, Codex, Claude Code and OpenClaw, each with its source on X.

Illustration: AI-generated (OpenAI image model via Codex): a radar screen with ten signals.
1.Hermes Agent now works in Microsoft Teams, the Microsoft Store and Azure
Confidence: high (press release read; Marketplace listing not opened)According to the Nous Research press release of November 18, Hermes Desktop is in the Microsoft Store, the agent works in Teams and its work runs in isolated Azure Container Apps microVMs. With Collective Wisdom you share a skill with your team after review. Companies with strict rules can run it all in their own Azure tenant.
Source on X: @NousResearch · Nous Research, persbericht
Context in the briefing: Hermes Enterprise in the November 20 briefing
Read also: Hermes Enterprise in Teams and Azure · Hermes Agent 0.23.2
2.GPT-Live-1.1 is in the API, with standby at 2 cents per minute
Confidence: high (forum post, model page and changelog read)The voice model behind ChatGPT voice since October 15 is now available as gpt-live-1.1. It listens while it speaks and hands heavy work to a model of your choice. New is standby: after 5 to 60 seconds of silence the model pauses for 0.02 dollars per minute, against 0.05 dollars active.
Source on X: @OpenAIDevs · OpenAI Developer Community
Context in the briefing: OpenAI in the November 20 briefing
Read also: GPT-Live-1.1 in the API · GPT-6.2 Astra and Sol
3.OpenAI has Codex agents find and fix regressions in ChatGPT
Confidence: medium (official account; no underlying page read)@OpenAIDevs describes the Performance Factory: Codex agents look for performance regressions in ChatGPT, propose a fix and then check in production whether it works. A useful pattern for your own agent teams: detect, propose, verify. We did not read an underlying page with figures.
Source on X: @OpenAIDevs
Context in the briefing: OpenAI in the November 20 briefing
Read also: GPT-6.2 Astra and Sol
4.Perplexity Brain writes its own skills: over 80,000 since August
Confidence: high for how it works (blog read); medium for the figures (from the maker)According to a Perplexity technical blog post, Brain builds personal skills from past sessions, with rules: a new skill only after the same behaviour shows up twice, one change per run and at most ten per user. In an eight-week test, failed tool calls fell 11 percent. Only for Max subscribers in the US.
Source on X: @perplexity_ai · Perplexity, Self-improving skills
Context in the briefing: Perplexity Brain in the November 20 briefing
Read also: Perplexity Brain and skills · Perplexity Teammates
5.Anthropic stock opened at 111.50 dollars and closed at 97.17 dollars
Confidence: medium (prices from a financial site via the X radar; we read no market feed)According to Stock Titan, read by the 09:00 X radar, the first trade was 111.50 dollars (34 percent above the 83-dollar offer price) and the close 97.17 dollars (17 percent higher). The 34 percent in posts on X belongs to the open, not the close. The valuation depends on which share count you use. Not investment advice.
Source on X: @kimmonismus · @AnthropicAI
Context in the briefing: The IPO in the November 20 briefing
Read also: Anthropic's S-1/A
6.Codex 0.171.0 makes GPT-6.2 Sol the default in the model catalog
Confidence: high (release notes on GitHub read)The stable release of November 20 adds GPT-6.2 Sol and Astra, with Sol as the default choice. The TUI workspace is now on by default, codex exec --remote starts a session on an existing app server, and /rc and /remote show pairing codes for remote control.
Source on X: @Codex_Changelog · GitHub, rust-v0.171.0
Read also: GPT-6.2 Astra and Sol
7.Claude Code: tokens double-counted in 2.1.328, config reset fixed in 2.1.329
Confidence: high for 2.1.329 (release notes read); low for the double count (one account)@StatsWire reports that 2.1.328 double-counted tokens and warns that two agents can keep talking to each other for hours and burn the weekly quota. According to its release notes, version 2.1.329 of November 19 fixes ~/.claude.json sometimes resetting to defaults after a network stall, and the trust dialog now says where a folder's hooks are declared.
Source on X: @StatsWire · GitHub, v2.1.329
Context in the briefing: Claude Code in the November 20 briefing
Read also: Claude Code 2.1.326
8.Cloudflare: Clairic out of preview, and a cf CLI for coding agents
Confidence: high for Clairic (official account, blog read earlier); medium for the cf CLI@Cloudflare says Clairic is out of preview and Clairic-flash is on Workers AI, both also as open weights. The cf CLI 1.0 RC is one command-line tool for the whole Cloudflare API, built with coding agents in mind. The latter is an earlier announcement making the rounds again.
Source on X: @Cloudflare · @CloudflareDev
Context in the briefing: Cloudflare in the November 20 briefing
Read also: Clairic open-weight
9.OpenClaw 2026.11.6: a patch with 386 pull requests and cache fixes
Confidence: medium (GitHub release read; changelog details via the X radar)The stable release of November 19 counts 406 commits, 386 pull requests and 157 contributors. According to the X radar it fixes, among other things, conversation history that broke prompt cache reuse. @steipete reports a fix for users of the embedded harness; whether it is in this tag is not confirmed.
Source on X: @steipete · GitHub, v2026.11.6
Context in the briefing: Unconfirmed in the November 20 briefing
10.Gemini 4 Helium is teased, but is not a release yet
Confidence: low (X only, teaser)@testingcatalog reports that Gemini 4 Helium, a Flash Lite model, is coming next. There is no model ID, no price and no changelog entry. The same account earlier guessed Neon would be a Flash Lite; Neon shipped as the middle model at 3 and 15 dollars. Treat Helium as a rumour.
Source on X: @testingcatalog
Context in the briefing: Unconfirmed in the November 20 briefing
Read also: Gemini 4 Neon


