Skip to content
NLEN

AI news, November 20, 2026: Anthropic listed, Hermes Enterprise, Perplexity Brain, GPT-Live-1.1

All briefings · Markdown version

Short answer: The main stories of November 19 to 20: Anthropic is publicly listed and publishes a statement from Dario Amodei, Hermes Agent reaches companies through Microsoft Teams and Azure, Perplexity Brain writes skills by itself and OpenAI brings GPT-Live-1.1 to the API. Earnings claims and paste-prompts around Grok Bot are unconfirmed.

Illustration: on the left a stock-exchange bell above a rising price chart, in the middle a team-app chat window with a robot assistant, on the right a gear drawing a skill card by itself.

Illustration by Claude (AI), drawn with code: a stock-exchange bell with a rising price (Anthropic listed), a team app with an agent (Hermes Enterprise) and a gear making a skill card (Perplexity Brain).

Anthropic is publicly listed; Dario Amodei publishes a statementConfidence: high for the listing and the price: the official account; low for the price after the open: one account, not checked.

Anthropic has been listed since November 19. Anthropic's account on X announced the price of 83 dollars a share (3.3 million views) and published a statement from Dario Amodei (1.5 million views). A third post from the account, containing only a link, reached 9.1 million views, more than eight times the account's median over the past 21 days.

@kimmonismus writes that the stock opened 34 percent above the offer price at a market value of 2 trillion dollars. That is one account's claim; we did not check the price ourselves and do not adopt the figure as fact.

What this means: the listing is confirmed by the company itself. Nothing reliable has been established about the price after the open. For people who use Claude nothing changes in the product; quarterly figures from a listed company will become a new source on revenue and costs. Background on the registration figures is in our article on Anthropic's S-1/A. This is not investment advice.

Sources: @AnthropicAI over de prijs van 83 dollar op X · @AnthropicAI: verklaring van Dario Amodei op X · @AnthropicAI: post met alleen een link op X · @kimmonismus over de opening van het aandeel op X

Hermes Agent comes to Microsoft Teams, the Microsoft Store and AzureConfidence: medium: official post with confirmation per the scout selection, not re-verified by us.

Nous Research announced Hermes Enterprise: Hermes Agent works in Microsoft Teams, is listed in the Microsoft Store and is available through Azure, with the option to run in your own Azure tenant (354 thousand views, over 1,500 bookmarks). The November 20 scout selection says the announcement is confirmed on the Nous page and in the Microsoft Marketplace; we did not reopen those pages ourselves in this round.

In addition, Qwen4-Max stays free for five days in the Nous Portal and in Hermes Agent, starting November 18. The original post by @NousResearch stands at 504 thousand views. The model is available only through the API and cannot be run locally.

What this means: the same agent framework gets a route to companies that use Microsoft services. Anyone who self-hosts Hermes does not need to change anything. Anyone who wants to test Qwen4-Max has until about November 23. We earlier wrote about Qwen4-Max and about Hermes Agent 0.23.2.

Sources: @NousResearch over Hermes Enterprise op X · @NousResearch over Qwen4-Max gratis op X · @0xdierich over Qwen4-Max in Hermes op X

Perplexity Brain writes skills from your own workflow; Teammates come to mobileConfidence: medium: official accounts; the figures come from the maker itself.

Perplexity reports that Brain in Perplexity Computer extracts repeated workflows from sessions and turns them into personal, self-improving skills (41 thousand views for Perplexity's account, 24.6 thousand for @AravSrinivas). According to Perplexity, 72.9 percent of users received a skill they did not write themselves, against 6.5 percent who added one themselves. Teammates and Multiagent are now also in the iOS and Android apps, with voice for Teammates (107 thousand views).

What this means: the percentages come from Perplexity itself and were not measured independently. The principle, skills from repeated work, resembles what Hermes already does; anyone who maintains their own skill library can compare the automatically made skills with it. We earlier wrote about Perplexity Teammates.

Sources: @perplexity_ai over Brain en skills op X · @AravSrinivas over de percentages op X · @perplexity_ai over Teammates op mobiel op X

OpenAI: GPT-Live-1.1 in the API and Codex agents that track down regressions in ChatGPTConfidence: high for GPT-Live-1.1 and the Performance Factory: official posts; medium for Dots: a news account and a user.

@OpenAIDevs reports GPT-Live-1.1 in the API (390 thousand views): more natural and consistent accents in all languages and lower cost during silence. A second post describes the Performance Factory (118 thousand views): Codex agents investigate performance regressions in ChatGPT, propose fixes and then check in production whether they work. According to @testingcatalog, Dots is rolling out to a small share of Plus users and to all Business Standard users; one user reports that it is not yet available in Europe.

What this means: the lower cost during silence makes voice agents cheaper, but the post gives no figure for the size of the reduction. The Performance Factory is a usable pattern for your own agent teams: detect, propose a fix and verify it. About Dots only the phased rollout is known. The GPT-6.2 Astra model is covered in our article on GPT-6.2 Astra and Sol.

Sources: @OpenAIDevs over GPT-Live-1.1 op X · @OpenAIDevs over de Performance Factory op X · @testingcatalog over Dots op X · @Riftv_ over Dots in Europa op X

Claude Code 2.1.328 fixes double token counting; warning about agent loopsConfidence: low to medium: the release is confirmed, the details come from one account with little reach.

@StatsWire reports that in Claude Code 2.1.328 tokens were counted twice and mentions fixes for subagents that are retried (1.9 thousand views). The November 19 X radar confirmed a new release on GitHub. The same account warns that two Claude Code agents can keep talking to each other for hours and use up the weekly quota (939 views). @claudeai thanked open-source builders (605 thousand views).

What this means: usage overviews from before this version may be too high. For overnight runs with several agents a budget and a stop limit help. The reports come from one account with little reach and were not reproduced by us. We earlier wrote about Claude Code 2.1.326.

Sources: @StatsWire over de tokentelling in 2.1.328 op X · @StatsWire over agent-loops op X · @claudeai bedankt open-source-bouwers op X

Cloudflare: Clairic out of preview, a cf CLI for agents and KitesurfConfidence: high for Clairic: the official account; medium for the cf CLI and Kitesurf: earlier announcements that are circulating again.

@Cloudflare reports that Clairic is out of preview, that Clairic-flash is on Workers AI and that both are released as open weights (9.4 thousand views). The cf CLI 1.0 RC is a single command-line program for the whole Cloudflare API, built with coding agents in mind (original by @CloudflareDev, 29 thousand views). Kitesurf, a browser engine for agents on Workers without a Chromium container, is open source under Apache 2.0; its changelog dates from November 12, so it is not news from today. Deploying it yourself requires the Workers Paid plan.

What this means: the Clairic models can therefore also be run locally, according to the X radar with 27 and 9 billion parameters under Apache-2.0; we did not check those details. Anyone considering Kitesurf as a replacement for a headless Chromium should read the code first. We earlier wrote about Clairic and about Kitesurf.

Sources: @Cloudflare over Clairic en Clairic-flash op X · @CloudflareDev over de cf-CLI op X · @CamilleRoux over Kitesurf op X

Unconfirmed and warnings: earnings claims, paste-prompts, 532 plugins and Gemini 4 HeliumConfidence: low: one or a few posts per claim, no primary confirmation.

One account reports that Gemini 4 Helium (Flash Lite) is teased as next (@testingcatalog, 62 thousand views); that is a teaser, not a release. OpenClaw user @joelniklaus reports that one setting for passing reasoning between agent steps raised the score by 44 percent (6.7 thousand views); that was not re-measured. @steipete reports a fix for users of OpenClaw's embedded harness (35 thousand views).

Warnings: one account claims to have earned 6,230 dollars with a Grok Bot guide and an Elon Musk repost (139 thousand views), a pattern of hype with earnings claims. Another shares a paste-prompt of unproven origin. A third writes that Anthropic has released 532 plugins for free (28 thousand views); the number and origin are not confirmed. A fourth claims to have sold a 100 thousand dollar agent to Anthropic (47 thousand views).

What this means: do not adopt paste-prompts, install lists or earnings claims without a primary source, and read the code of unknown plugins before installing them. We earlier wrote about OpenClaw in our article on the approval step in OpenClaw 2026.11.4.

Sources: @testingcatalog over Gemini 4 Helium op X · @joelniklaus over de reasoning-instelling in OpenClaw op X · @steipete over de fix voor de embedded harness op X · @Bober_smart over de verdienclaim op X · @Apexxxzz over 532 plugins op X

Primary sources: Nous Research: Hermes Enterprise (X) · Cloudflare: Clairic en Clairic-flash (X) · OpenAI Developers: GPT-Live-1.1 (X)