The AI Gazette · Morning edition · Tuesday 20 October 2026luongnv.com/news/reports/2026-10-20-morning.html

Morning edition — Tuesday 20 October 2026

Reporting window: to (Paris time). Issued .

This edition follows the Monday afternoon issue, published at 16:09 CEST on 19 October. Five verified developments came after it: Inception's Mercury 3 diffusion model, reported talks on a second roughly $45 billion compute deal between Nscale and Anthropic, Reuters' report that OpenAI is accelerating a roughly $30 billion raise before Anthropic's IPO, Germany's declaration of intent with Nvidia on industrial AI, and a judge's order keeping the former OpenAI engineer accused of bugging an executive conference room in custody. One older item appears under BACKGROUND with its original date: the 15 October preprint on Gemini agents completing Minecraft, which the previous edition held back until the paper could be found. Stories from earlier editions are not repeated.

5 new or updated dispatches · 1 background item · Previous: 2026-10-19-afternoon

The News

Models New

Inception releases Mercury 3, a diffusion language model it says runs at more than 1,000 tokens per second

Published:

Inception announced Mercury 3 on 19 October and called it the most capable diffusion-based language model released to date. Diffusion models generate many tokens in parallel through repeated denoising, rather than one token at a time.

The company says Mercury 3 is more than 40 times larger than Mercury 1, runs at 1,068 tokens per second on Nvidia GPUs and has a one-million-token context window, up from 260,000 for Mercury 2.5. It claims quality on par with or better than speed-optimised models such as GPT-6 Luna and Claude Haiku 5.5, while running four to eight times faster.

List prices are $0.20 per million input tokens and $0.70 per million output tokens, with a 50% launch discount to $0.10 and $0.35. The model is available through Inception's own API, OpenRouter, Baseten and Vercel, and the API is compatible with OpenAI's.

Caveat. All benchmark and speed figures come from Inception, and this edition found no independent test. Inception has not published the parameter count or said whether it will release the weights. Both questions were asked in the launch thread.

Sources

  1. Inception (primary source) Checked:
  2. Business Wire via Yahoo Finance (primary source) Checked:
  3. Hacker News launch thread (community) Checked:

Industry & policy New

Nscale and Anthropic are in talks on a second compute deal worth about $45 billion, Bloomberg reports

Published:

Bloomberg News reported on 19 October, citing people familiar with the negotiations, that the Nvidia-backed data-centre developer Nscale is in talks to sell Anthropic computing power in a deal that could be worth about $45 billion. The deal would cover a second facility at Nscale's Monarch campus in West Virginia.

According to Nscale's IPO filing, Anthropic already has agreements with Nscale worth as much as about $44.6 billion. As relayed by Reuters, Bloomberg said a second Monarch building would deliver 460 megawatts using Nvidia's Vera Rubin chips, and that Anthropic holds a right of first refusal on a third building. The sources said talks are continuing and the details could change.

Reuters said it could not independently verify the report, and neither company had commented when TechCrunch and Reuters published. The talks follow Microsoft's withdrawal from a separate Nscale data-centre deal, reported in this journal's 19 October morning edition.

Caveat. This is an unconfirmed report of negotiations based on anonymous sources, not an announced agreement.

Sources

  1. Bloomberg News via Bloomberg Government (reporting) Checked:
  2. Reuters via CNA (reporting) Checked:
  3. TechCrunch (reporting) Checked:

Industry & policy New

OpenAI is racing to raise about $30 billion before Anthropic's expected IPO, Reuters reports

Published:

Reuters reported late on 19 October, citing two people familiar with the matter, that OpenAI is trying to secure investor commitments before Anthropic's expected stock-market debut in mid-November. The round could value OpenAI at more than $1.4 trillion. According to one source, SoftBank, Abu Dhabi's MGX and other existing backers are in talks to join a round of about $30 billion.

One source said OpenAI is telling investors it reached $50 billion in annualised revenue after launching GPT-6 Astra, and that it expects $70 billion by the end of the year. OpenAI was last valued at $852 billion in March, after a $122 billion round.

According to Reuters, OpenAI and SoftBank declined to comment, and Anthropic and MGX did not immediately respond.

Caveat. The revenue figures are company claims to investors, relayed by an anonymous source. The round size and valuation are not final.

Sources

  1. Reuters via CNA (reporting) Checked:

Industry & policy New

Germany signs a joint declaration of intent with Nvidia on industrial and physical AI

Published:

Nvidia said that on 19 October, at the Invest in Germany Summit in Berlin, Chancellor Friedrich Merz, the Federal Ministry for Economic Affairs and Energy and the Federal Ministry for Research, Technology and Space signed a Joint Declaration of Intent with the company. It covers a joint initiative to advance AI models for physical AI and formalises Nvidia's work with the government's Industry 4.0 Platform.

Nvidia chief executive Jensen Huang also announced planned investments in four German startups: Agile Robots, Avelios Medical, Mirelo and Reflex Aerospace. Nvidia said these are part of a commitment of nearly $10 billion across Europe.

Merz urged small and medium-sized companies to adopt AI: "Please open your doors for AI to make use of your data," he said, according to Nvidia's account of the event.

Caveat. This account comes from Nvidia's blog. The text of the declaration and the size of each investment were not published there.

Sources

  1. NVIDIA Blog (primary source) Checked:

Industry & policy New

Judge orders former OpenAI engineer held without bail in conference-room recording case

Published:

At a detention hearing on Monday 19 October, Judge Alex Tse of the federal court in San Francisco ordered Abdulrahman Alfozan, 35, held without bail pending trial, saying "there is a risk of non-appearance that is significant". Alfozan, a Saudi national and former OpenAI engineer, is charged with trying to intercept oral communications. Prosecutors say that on or about 28 September he hid a recording device in a television in an executive conference room.

In a detention memorandum, prosecutors said Alfozan was charged on 8 October, arrested on 9 October and fired by OpenAI. They said he had been in frequent WhatsApp contact with a member of the Saudi royal family, and that a Saudi military official paid for one-way business-class tickets for his wife and son, who left the United States the day after his arrest. They called the case a national-security concern and said more serious charges may follow.

CNBC reported that the judge called an offer by the Saudi consulate to pay the full bail "problematic". Alfozan's lawyer argued that prosecutors had not shown he was a serious flight risk. The charge carries a maximum of five years in prison, and Alfozan is due back in court on 6 November.

Caveat. The allegations are unproven, and the complaint does not charge espionage. The memorandum states the prosecution's position.

Sources

  1. US Attorney's detention memorandum (CourtListener) (primary source) Checked:
  2. BBC News (reporting) Checked:
  3. CNBC (reporting) Checked:

Background — older or undated context

Earlier or undated items, carried for context only. They are not new reporting.

Research Background — older context

BACKGROUND: Four Gemini 4 Argon agents finish Minecraft in 27 hours, a 15 October preprint reports

Published:

This preprint was submitted to arXiv on 15 October. It drew wide coverage from 19 October, and the previous edition left it out because the paper had not yet been found. The authors, Maxence Faldor, Satinder Singh and Roberta Raileanu, placed four Gemini 4 Argon agents in a sandboxed Minecraft survival world on Easy difficulty and gave them one goal: defeat the Ender Dragon.

The agents used the open-source Mineflayer library, coordinated through in-game chat and shared memory, and wrote and debugged their own programs without assigned roles. The authors report that the team wrote 2,783 programs and died 605 times before defeating the dragon after 27 hours and 14 minutes. They describe it as, to their knowledge, the first end-to-end autonomous completion of Minecraft by AI agents without pre-scouted seeds or game-specific human-written tools.

Caveat. This is a single run reported by its authors in an unreviewed preprint, on Easy difficulty and with researchers able to intervene. It has not been independently replicated.

Sources

  1. arXiv 2610.22319 (primary source) Checked:

Key patterns — editorial analysis

  • The capital race between the frontier labs is speeding up. OpenAI is reportedly timing a roughly $30 billion raise around Anthropic's IPO, and Anthropic is reportedly negotiating another compute contract of about $45 billion.
  • Speed and price are now selling points of their own: Inception is pitching diffusion models at more than 1,000 tokens per second for agent loops.
  • Security and sovereignty concerns are reaching AI labs and their host countries: a US court held an alleged insider who bugged an OpenAI executive room, while Germany signed an industrial-AI pact with Nvidia.

Source coverage & colophon

Collection engine

Version
3.28.0
Ran this edition
yes
Engine look-back
2 days

Source status

reddit
ok
youtube
ok
hackernews
ok
polymarket
unreachable
github
ok
devto
not searched

Coverage notes

  • The fresh window runs from the previous issue (16:09 CEST, 19 October) to this issue. The court story carries a date-only stamp; the hearing took place on 19 October, after the previous issue.
  • The installed last30days engine completed with 73 evidence items from Reddit, Hacker News, GitHub and YouTube. Three web supplements followed, plus direct checks of primary sources and Google News RSS pre-research. Evidence-item counts are not verified-development counts. Community quotes come from the Mercury 3 launch thread on Hacker News and are anecdotal.
  • Polymarket failed its TLS certificate hostname check and verification was not bypassed. YouTube returned items, but transcript downloads hit HTTP 429 rate limits. dev.to was not in this run's query-plan sources and returned no articles. X was not searched. Failed or missing coverage is not read as an absence of discussion.
  • The configured web search backend failed and its keyless fallback served the searches. Public HTTPS pages were fetched directly with normal certificate validation. OpenAI's website returned HTTP 403. Reuters facts come from outlets that republish or cite it.
  • Excluded after date checks: Gemini 4 Argon (13 October), Nvidia's Nemotron 3.5 Super VL and Bolt models (15 October), Microsoft's agentic-insider controls and OpenAI's COBOL and low-resource-language evaluations (both published on 19 October before the previous issue; OpenAI's page could not be checked), Amazon's Indiana nuclear plan (19 October, before the previous issue), the Canadian publishers' copyright suit against Anthropic (filed 6 October, reported 14 October), Meta's revised scraping terms (16 October), and new reactions to Anthropic's draft AI bill (already covered).
  • No browser-cookie access, optional service or credential signup, paid enrichment, TLS bypass, Job Alerts edits or schedule changes were used. Research logs and full provenance remain private.

Engine footer