2026-10-30 notis CATEGORY Join
Show Read as Find Sections Models and releasesAgents and appsResearchMoney and computeLaws and rulingsSafety and securityAI at work More ArchiveModelsAboutCorrectionsThe numbersRSS
My notis
TODAY'S NEWS
October 2026 3 STORIES
MTWTFSS

Days with a dot have news. See the archive.

The whole week ›
17:30
0/3
WEATHER THEME CARD SPACING

Today's news

today

My page

Kept in this browser only.

SAVED FOLLOWING COLLECTIONS READING SINCE
Nothing saved yet. Tap the bookmark on a card to keep it here. read today
all cases

No case followed yet.

Price changes Stories where a published price went up or down.
Models being retired Models being withdrawn, and gaps before a replacement is ready.
The day's top stories. Sign up
toned down Models 30 OCT

Mistral releases Large 4 weights, but paid hosting needs a separate licence

LicenceMistralMistral LargeOpen weights
Illustration: A white cabinet with an open green hatch and slide-out cartridges, and a latched gate at its side, standing for released weights and a hosting licence.Illustration made with AI for ai notis

Mistral says Large 4, a 1-trillion-parameter multimodal model, is open-weight under a modified MIT licence; a business selling it as a service must take a separate licence.

SOURCEMistral AI · 3 sources
CHECKED30 Oct, 05:21
Models read
Illustration: A white cabinet with an open green hatch and slide-out cartridges, and a latched gate at its side, standing for released weights and a hosting licence. Models / 30 OCT

Mistral releases Large 4 weights, but paid hosting needs a separate licence

  • Mistral says it is releasing the weights of Mistral Large 4, a 1-trillion-parameter, natively multimodal mixture-of-experts model with 49B active parameters per token.
  • Use, modification, fine-tuning and commercialisation are allowed, but a business that operates a Model-as-a-Service offering must get a separate commercial licence from Mistral.
  • Mistral lists API prices of $0.70 per million input tokens, $0.07 per million for cached input and $2.10 per million output tokens.

Mistral AI, whose co-founders say they wanted a European leader, says it is releasing the weights of Mistral Large 4, which it describes as a 1-trillion-parameter, natively multimodal mixture-of-experts model. The model card lists 49B active parameters per token, a 1M-token context length and 128 routed experts with 4 active per token. The licence is the Mistral Ecosystem Open-Weight License, a modified MIT licence. Use, modification, fine-tuning and commercialisation are allowed, but a business that runs a Model-as-a-Service offering must get a separate commercial licence from Mistral. Mistral says it trained the model from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its European datacentres. Mistral lists API prices of $0.70 per million input tokens and $2.10 per million output tokens. Mistral says it scores 48% on the Artificial Analysis Coding Agent Index, roughly tied with GLM-5.3 and using about 47% fewer tokens per task. These benchmark results are Mistral's own account. On cyber, Mistral says it measures a 96.6% average refusal rate on malicious cyber requests and also says it sees potential for uplift to conduct real attacks, so it is changing its guardrails. A cybersecurity variant is in preview for vetted design partners.

What it means for you Our view

If you plan to host Large 4 for customers, read the Mistral Ecosystem Open-Weight License before you commit. Mistral says a business that operates a Model-as-a-Service offering needs a separate commercial licence, while use, modification and fine-tuning are allowed. So the licence terms depend on how you use the weights as well as on the download. The benchmark results, including the 48% Mistral reports on the Artificial Analysis Coding Agent Index, are Mistral's own account and worth testing. Before you switch, run Large 4 on your own tasks and compare its listed prices of $0.70 and $2.10 per million tokens with what you pay now.

TAGSLicenceMistralMistral LargeOpen weights
Compute 30 OCT

Meta prices $25bn stock sale as Nvidia and AMD cite Muse AI agents

AMDMetaMuseNvidiaVera Rubin
Illustration: A steel press with two intake chutes and one output tray; the green chute stands for the upsized common stock sale, the second for preferred-linked shares.Illustration made with AI for ai notis

Meta priced a $25.0bn Class A stock sale on 26 October; Nvidia says it will invest $3bn and AMD $1bn, and their releases describe Meta's AI agent and Muse work.

SOURCESEC EDGAR · 5 sources
CHECKED30 Oct, 05:41
Compute read
Illustration: A steel press with two intake chutes and one output tray; the green chute stands for the upsized common stock sale, the second for preferred-linked shares. Compute / 30 OCT

Meta prices $25bn stock sale as Nvidia and AMD cite Muse AI agents

  • Meta priced Class A stock upsized to $25.0bn from $22.5bn at $625.00 a share, and 250,000,000 depositary shares at $50.00 each, on 26 October.
  • Nvidia's release says it will invest $3bn in the offering and AMD's says $1bn; Meta's filing says Nvidia, AMD and Broadcom's interest, up to $5.0bn in total, is not binding.
  • Nvidia says Meta intends to expand its Vera Rubin and Vera CPU deployment for next-generation AI agents; AMD says EPYC processors support the environments behind Meta's Muse agent.

Meta Platforms priced two concurrent offerings on 26 October: Class A common stock and depositary shares in a 6.50% mandatory convertible preferred stock. The common stock offering was upsized to $25.0bn from $22.5bn, at $625.00 a share. The 250 million depositary shares are priced at $50 each. Meta's pricing term sheet gives expected net proceeds of about $12.5bn from the depositary shares and about $24.9bn from the common stock, which is sold through forward sale agreements and is expected to settle by 19 November 2027. Meta intends to use the proceeds for general corporate purposes. Its prospectus supplement says NVIDIA, AMD and Broadcom have indicated interest in buying up to $5.0bn of the Class A stock in total, and that these indications are not binding. In its own release, NVIDIA says it will invest $3bn in Meta's equity offering; AMD's release says it will invest $1bn. NVIDIA's release gives no share count, price or valuation. NVIDIA's release also says Meta intends to expand its deployment of Vera Rubin and Vera CPUs for next-generation AI agents, and that the two companies will expand their work on Muse models; AMD's release says its EPYC processors support the computing environments behind Meta's Muse personal AI agent. Meta separately says it plans an at-the-market programme of up to $60bn of Class C common stock, not expected to start before the first quarter of 2027, with no agreement signed.

What it means for you Our view

This is a financing story with an AI hardware thread, and that thread rests on the chipmakers' own releases. Meta's term sheet says it intends to use the net proceeds from the depositary shares for general corporate purposes, so the filings do not tie the money to Vera Rubin or Muse. Nvidia's and AMD's investments are stated in their releases, while Meta's prospectus calls the indications of interest non-binding. If you buy or plan capacity from Nvidia or AMD, treat the Vera Rubin and EPYC mentions as signals about Meta's suppliers, and read Meta's next filing for what the money funds.

TAGSAMDMetaMuseNvidiaVera Rubin
SEC EDGAR · 5 sources tap to close
Compute 30 OCT

Oracle says Nvidia Vera Rubin NVL72 racks are generally available, price not given

NvidiaOpenAIOracleOracle Cloud InfrastructureVera Rubin
Illustration: A white utility cabinet with a green bank of drawers and an empty plate beside the door, standing for racks now available and no price stated.Illustration made with AI for ai notis

Oracle and Nvidia say Vera Rubin NVL72 racks and Vera CPUs are available on OCI now, and OpenAI says it is bringing Vera into production there for its agentic sandbox workloads.

SOURCENVIDIA · 1 source
CHECKED30 Oct, 05:41
Compute read
Illustration: A white utility cabinet with a green bank of drawers and an empty plate beside the door, standing for racks now available and no price stated. Compute / 30 OCT

Oracle says Nvidia Vera Rubin NVL72 racks are generally available, price not given

  • Oracle's and Nvidia's joint release of 27 October says Vera Rubin NVL72 and Vera CPUs are generally available on Oracle Cloud Infrastructure; each rack connects 72 GPUs through NVLink.
  • Uday Ruddarraju, chief technology officer of compute at OpenAI, is quoted saying OpenAI is bringing Vera into production on OCI for its agentic sandbox workloads.
  • The release states no price. Expanding OCI Superclusters to multi-gigawatt scale and adding HGX Rubin NVL8 systems with eight Rubin GPUs are plans only.

Oracle and NVIDIA said on 27 October that NVIDIA Vera Rubin NVL72 and NVIDIA Vera CPUs are generally available on Oracle Cloud Infrastructure. Each NVL72 rack connects 72 GPUs through NVLink. The companies separately announced plans to expand OCI Superclusters with Vera Rubin NVL72 to multi-gigawatt scale and to introduce HGX Rubin NVL8 systems, which are plans, not available now. OpenAI says it is bringing Vera CPUs into production on OCI for its agentic sandbox workloads. The release gives no price.

What it means for you Our view

If you plan AI capacity on Oracle Cloud Infrastructure, you can now ask for Vera Rubin NVL72 racks and Vera CPUs, but the release states no price, so cost and capacity terms have to come from Oracle. The OpenAI quote concerns Vera CPUs for its agentic sandbox workloads, not the racks. The multi-gigawatt expansion and the HGX Rubin NVL8 systems are plans in the release, not products you can order today. Before you design around either, ask Oracle for pricing and delivery terms in writing.

TAGSNvidiaOpenAIOracleOracle Cloud InfrastructureVera Rubin
NVIDIA · 1 source tap to close
NOTHING IN THIS FILTER No story today matches. open the archive →
That's today's news: three stories. ExplainedHow speculative decoding lets a small model draft and a large one check

The morning email

On the mornings we publish: the three top stories and up to four short ones. Free.

We email you a link to confirm. An issue may include one sponsor, always labelled Sponsored · Advertisement. Our emails count opens and clicks, not who made them. Unsubscribe in one click. What we keep

1 of 3
Illustration: A white cabinet with an open green hatch and slide-out cartridges, and a latched gate at its side, standing for released weights and a hosting licence. Illustration made with AI for ai notis

Models · 30 OCTnarrowed

Mistral releases Large 4 weights, but paid hosting needs a separate licence

Mistral says Large 4, a 1-trillion-parameter multimodal model, is open-weight under a modified MIT licence; a business selling it as a service must take a separate licence.

  • Mistral says it is releasing the weights of Mistral Large 4, a 1-trillion-parameter, natively multimodal mixture-of-experts model with 49B active parameters per token.
  • Use, modification, fine-tuning and commercialisation are allowed, but a business that operates a Model-as-a-Service offering must get a separate commercial licence from Mistral.
  • Mistral lists API prices of $0.70 per million input tokens, $0.07 per million for cached input and $2.10 per million output tokens.
Checked 30 Oct, 05:21 · Mistral AI · 3 sources
Illustration: A steel press with two intake chutes and one output tray; the green chute stands for the upsized common stock sale, the second for preferred-linked shares. Illustration made with AI for ai notis

Compute · 30 OCT

Meta prices $25bn stock sale as Nvidia and AMD cite Muse AI agents

Meta priced a $25.0bn Class A stock sale on 26 October; Nvidia says it will invest $3bn and AMD $1bn, and their releases describe Meta's AI agent and Muse work.

  • Meta priced Class A stock upsized to $25.0bn from $22.5bn at $625.00 a share, and 250,000,000 depositary shares at $50.00 each, on 26 October.
  • Nvidia's release says it will invest $3bn in the offering and AMD's says $1bn; Meta's filing says Nvidia, AMD and Broadcom's interest, up to $5.0bn in total, is not binding.
  • Nvidia says Meta intends to expand its Vera Rubin and Vera CPU deployment for next-generation AI agents; AMD says EPYC processors support the environments behind Meta's Muse agent.
Checked 30 Oct, 05:41 · SEC EDGAR · 5 sources
Illustration: A white utility cabinet with a green bank of drawers and an empty plate beside the door, standing for racks now available and no price stated. Illustration made with AI for ai notis

Compute · 30 OCT

Oracle says Nvidia Vera Rubin NVL72 racks are generally available, price not given

Oracle and Nvidia say Vera Rubin NVL72 racks and Vera CPUs are available on OCI now, and OpenAI says it is bringing Vera into production there for its agentic sandbox workloads.

  • Oracle's and Nvidia's joint release of 27 October says Vera Rubin NVL72 and Vera CPUs are generally available on Oracle Cloud Infrastructure; each rack connects 72 GPUs through NVLink.
  • Uday Ruddarraju, chief technology officer of compute at OpenAI, is quoted saying OpenAI is bringing Vera into production on OCI for its agentic sandbox workloads.
  • The release states no price. Expanding OCI Superclusters to multi-gigawatt scale and adding HGX Rubin NVL8 systems with eight Rubin GPUs are plans only.
Checked 30 Oct, 05:41 · NVIDIA · 1 source

You're all caught up

That's today's news: three stories.

ExplainedHow speculative decoding lets a small model draft and a large one check

You can join ai notis soon, to follow cases and keep stories on every device.

Follow us on Instagram and Threads

The morning email

On the mornings we publish: the three top stories and up to four short ones. Free.

We email you a link to confirm. An issue may include one sponsor, always labelled Sponsored · Advertisement. Our emails count opens and clicks, not who made them. Unsubscribe in one click. What we keep

No story today matches. Open the archive
Today's stories 3 STORIES

Today's stories

Fri 30 Oct · 3 STORIES
  1. Models · 30 OCT · narrowed

    Mistral releases Large 4 weights, but paid hosting needs a separate licence

    Illustration: A white cabinet with an open green hatch and slide-out cartridges, and a latched gate at its side, standing for released weights and a hosting licence.Illustration made with AI for ai notis

    Mistral says Large 4, a 1-trillion-parameter multimodal model, is open-weight under a modified MIT licence; a business selling it as a service must take a separate licence.

    Checked 30 Oct, 05:21 · Mistral AI · 3 sourcesRead

  2. Compute · 30 OCT

    Meta prices $25bn stock sale as Nvidia and AMD cite Muse AI agents

    Illustration: A steel press with two intake chutes and one output tray; the green chute stands for the upsized common stock sale, the second for preferred-linked shares.Illustration made with AI for ai notis

    Meta priced a $25.0bn Class A stock sale on 26 October; Nvidia says it will invest $3bn and AMD $1bn, and their releases describe Meta's AI agent and Muse work.

    Checked 30 Oct, 05:41 · SEC EDGAR · 5 sourcesRead

  3. Compute · 30 OCT

    Oracle says Nvidia Vera Rubin NVL72 racks are generally available, price not given

    Illustration: A white utility cabinet with a green bank of drawers and an empty plate beside the door, standing for racks now available and no price stated.Illustration made with AI for ai notis

    Oracle and Nvidia say Vera Rubin NVL72 racks and Vera CPUs are available on OCI now, and OpenAI says it is bringing Vera into production there for its agentic sandbox workloads.

    Checked 30 Oct, 05:41 · NVIDIA · 1 sourceRead

toned down Models 30 OCT

Mistral releases Large 4 weights, but paid hosting needs a separate licence

Mistral says Large 4, a 1-trillion-parameter multimodal model, is open-weight under a modified MIT licence; a business selling it as a service must take a separate licence.

Mistral AI · 3 sources·checked 30 Oct, 05:21
LicenceMistralMistral LargeOpen weights
Illustration: A white cabinet with an open green hatch and slide-out cartridges, and a latched gate at its side, standing for released weights and a hosting licence.Illustration made with AI for ai notis
Compute

Meta prices $25bn stock sale as Nvidia and AMD cite Muse AI agents

Meta priced a $25.0bn Class A stock sale on 26 October; Nvidia says it will invest $3bn and AMD $1bn, and their releases describe Meta's AI agent and Muse work.

30 OCT · SEC EDGAR · 5 sources
Compute

Oracle says Nvidia Vera Rubin NVL72 racks are generally available, price not given

Oracle and Nvidia say Vera Rubin NVL72 racks and Vera CPUs are available on OCI now, and OpenAI says it is bringing Vera into production there for its agentic sandbox workloads.

30 OCT · NVIDIA · 1 source

Explained

One idea behind the news
Research How speculative decoding lets a small model draft and a large one check A cheaper model guesses several tokens and a large one checks them in one pass; Mistral's Large 4 model card describes a draft head that does this.
In brief 3 more today
0 of 3 read
Drag a card to start Drag left or right · tap to read more Hover to lift a card · tap to bring it forward Tap again to put it back Tap a card to open it Tap again to close it Tap a headline to read it as a card