Friday, 16 October 2026

What actually happened in AI. Not what got the clicks.

Every story is boiled down to the event underneath it, then scored on evidence, concreteness, impact and actionability. Hype is penalised. A story needs evidence and specifics to reach the front page.

Live signals: 60
Avg score: 67
Screened, 24h: 84
Scored, 24h: 22

Last run 18:07 UTC. Most of what we read never gets here.

Newsletter

Get the no-hype AI news straight to your inbox

Weekly. No spam, unsubscribe any time. Privacy

Top signal

Meta releases Muse Spark Open, an open-weight ~816B-parameter multimodal mixture-of-experts model, with Hugging Face support

Meta published open weights for Muse Spark Open (~816B total, ~42B active per token, 256 experts with 8 routed per token, up to 1M-token context). It takes text, image, audio and video input and outputs text. It is tuned for agentic coding. Checkpoints are on Hugging Face in BF16 (~1.5 TB) and…

1d ago/4 sources/high confidence

CapabilityAccessInfrastructureAdoption

80Strong signal

Ranked feed60 live signals

China's commerce and science ministries publish draft export-control list covering 23 AI technologies, including large models above 10^26 FLOPs, 2 trillion parameters or global top-10 rank

China's commerce and science ministries released a draft update to the catalog of banned and restricted export technologies. It covers 23 AI areas across large models, agents, compute clusters and data processing. Autonomous malware-writing tools aimed at critical infrastructure, the military or…

5h ago/1 source/medium confidence

GovernancePowerAccess

Reflection AI releases open-weight Beam coding model (501B total, 23B active parameters) under Apache 2.0, available on AWS SageMaker JumpStart

Reflection AI released Beam 1.0, a sparse mixture-of-experts model for agentic coding (501B total, 23B active parameters, reported effective 1M-token context). Weights are on Hugging Face under Apache 2.0 in BF16, FP8 and NVFP4 variants. At launch it is deployable via Amazon SageMaker JumpStart in…

7h ago/2 sources/high confidence

CapabilityAccessInfrastructure

Anthropic gives six METR researchers employee-level access to assess its models, safety practices and risk management

Six METR researchers now have Anthropic-issued laptops, office space and internal tool access roughly on par with Anthropic's own risk assessment teams. METR will publish its own assessments of Anthropic's risks, incidents and practices. Anthropic cannot veto publication but can require redaction…

1d ago/2 sources/high confidence

GovernancePowerCapability

Black Forest Labs releases open FLUX.3 Image Klein models with day-one Diffusers support and bounding-box layout control

Two open checkpoints, a distilled one (flux-3-image-klein: 16 steps, no CFG) and a base one (flux-3-image-klein-base: 50 steps, CFG), are available on Hugging Face and run in Diffusers from day one as a Modular Diffusers pipeline. One model handles text-to-image, multi-reference image-to-image and…

1d ago/1 source/high confidence

CapabilityAccessAdoption

Character.AI files for Chapter 11 bankruptcy and proposes selling assets to Disney as lead bidder for $20 million

Character.AI filed for Chapter 11 bankruptcy protection in the US, listing estimated liabilities of $50-100 million. It proposes a sale process in which Disney is the lead (stalking-horse) bidder, paying $20 million for certain undisclosed assets, providing transition support using former Character…

3d ago/2 sources/high confidence

EconomicsPowerLabour

Microsoft releases open-source Probabilistic Bias Correction tool, published in Nature, to improve 2-6 week weather forecasts

Microsoft Research and collaborators (MIT, ECMWF, others) published a Nature paper on Probabilistic Bias Correction (PBC), a machine learning method that corrects systematic bias in existing weather models, and released it as open-source software. It runs on CPUs. In the study, PBC improved ECMWF's…

2d ago/1 source/high confidence

CapabilityEconomicsAccess

TheoPhysEval benchmark released: GPT-6-Astra in Codex solves 49 of 50 unpublished theoretical physics problems at about $116 total

Thirty-five researchers (ICTS, Harvard, KU Leuven, Google DeepMind and others) released TheoPhysEval, a benchmark of 50 unpublished theoretical physics research problems with verifiable answers across eleven subfields, with problems and test setup downloadable on Hugging Face. Results: GPT-6-Astra…

1d ago/1 source/high confidence

CapabilityEconomicsLabour

Google DeepMind's Gemini 4 Argon scores 97.5% on ARC-AGI-3 under vendor-specific harness, 49.8% under standard harness

ARC Prize reported Gemini 4 Argon scoring 97.5% on ARC-AGI-3 ($4,800) with the Provider Adapter Harness, making Google the second lab after OpenAI (GPT-6 Astra, 99.9%) to pass 95%. In the Standard Harness it scores 49.8% at low reasoning ($5,800), with API problems preventing completion of all 55…

8h ago/1 source/medium confidence

CapabilityEconomics

Google releases Gemini 4 Argon model globally via API, Antigravity, Gemini app and AI Mode in Search at $2/$10 per million tokens

Google made Gemini 4 Argon generally available on 13 Oct 2026: for developers and enterprises via the API in AI Studio and the Gemini platform; for Google AI Ultra subscribers in Antigravity; and for Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search. Introductory pricing…

3d ago/2 sources/high confidence

CapabilityEconomicsAccess

Meta to update commercial terms on October 30 requiring business accounts to let it crawl linked website content, including for training

Meta's consolidated Commercial Terms take effect October 30. Business accounts must authorize Meta to use automated crawlers and "test credentials" to access, index, cache and collect content from domains associated with their products or shared through Meta services, with no opt-out in the terms.…

7h ago/2 sources/medium confidence

GovernanceEconomicsPower

40 state attorneys general file notice of interest in Character.AI's Chapter 11 case, seeking data and child-safety protections in any asset sale

A bipartisan group of 40 state AGs, led by New York AG Letitia James, filed a notice of interest in Character.AI's bankruptcy case (Chapter 11, announced Tuesday along with a sale of some assets to Disney). They put parties on notice that regulatory issues must be addressed in any sale. They urge…

2d ago/1 source/high confidence

GovernanceEconomicsAdoption

NVIDIA and AI4Bharat announce Nemotron-pretraining-Indic-v1, a ~2.8T-token synthetic corpus across 22 Indian languages, with open generation pipeline to follow

NVIDIA, with IIT Madras' AI4Bharat, announced Nemotron-pretraining-Indic-v1: about 2.8T tokens (2,601.0B native-script, 224.6B romanized) covering 22 Indian languages in 34 language-script configurations. It is machine-translated from 15 Nemotron-CC-v2 English subsets using NeMo Data Designer and…

21h ago/1 source/high confidence

CapabilityAccessInfrastructure

Anthropic releases Claude Haiku 5.5 at sharply lower prices, halves Sonnet 5.5 cache read cost, and adds monthly API credits for subscribers

Anthropic released Claude Haiku 5.5, its small model, available on AWS, Google Cloud and Azure. It is the first Haiku with adjustable reasoning levels. Pricing per 1M tokens for prompts up to 100k is $0.10 input and $0.50 output (Haiku 4.5: $1 and $5). Prompts over 100k cost 5x as much. Anthropic…

1w ago/3 sources/medium confidence

EconomicsCapabilityAccess

Meta releases Muse Wonder and Muse Wonder Full models, scoring 52 and 56 on the Artificial Analysis Intelligence Index at roughly triple the per-task cost of Muse Spark 1.3

Meta released two frontier models. Muse Wonder is available via Meta's API and Muse Code. Muse Wonder Full is limited to Muse Code users. Both have a 1M-token context window and accept text, image, video and audio input. They add "Active Memory" and "Active Focus" features. Muse Wonder will soon…

4d ago/1 source/high confidence

CapabilityEconomicsAccess

arXiv limits each person to two paper submissions per month from October 1, 2026, citing AI-driven submission surge

Starting October 1, 2026, arXiv allows only two submissions per submitter per calendar month across all categories (co-authors unaffected). Rejected papers count toward the limit; deleting a paper before publication restores the slot. The existing cap of three active submissions remains. arXiv…

5d ago/1 source/high confidence

GovernanceAdoptionLabour

Nvidia releases open Nemotron 3.5 Super VL (120B MoE, 12B active) multimodal agent model, plus two smaller Bolt models for local hardware

Nvidia published open weights, datasets and training recipes for Nemotron 3.5 Super VL, a 120B-parameter Mixture-of-Experts multimodal model (12B active per token) distilled from Nemotron 3 Ultra and built for long-running enterprise agents and document/vision tasks. It also released Nemotron 3.5…

1d ago/1 source/medium confidence

CapabilityInfrastructureAccess

Anthropic tells investors Q3 revenue was $16.8bn, up nearly 50% from Q2, ahead of a planned IPO, per FT

According to anonymous sources cited by the FT, Anthropic told investors at an event on Wednesday that revenue was about $4.8bn in Q1, $11.5bn in Q2 and $16.8bn in Q3 2026. In Q3 it earned $1.70 in revenue per $1 spent on computing power, and its adjusted operating margin excluding stock-based…

1d ago/1 source/medium confidence

EconomicsAdoptionInfrastructure

Black Forest Labs releases FLUX 3 Image Klein, a 5.6B-parameter open-weights image model that runs on 8 GB GPUs

BFL released FLUX 3 Image Klein (5.6B parameters) with open weights under the FLUX Kommunity License (free commercial use for companies under $5M annual revenue). It supports text-to-image, editing, up to ten reference images, native 4K output and bounding-box placement. An INT8 build runs on…

1d ago/1 source/medium confidence

CapabilityAccessEconomics

Mistral releases Timestral, a family of four time-series foundation models (5M–209M parameters) claiming top ranks on GIFT-Eval, TIME and fev-bench

Mistral AI released Timestral, its first time-series foundation model family, in four sizes (5M, 28M, 123M, 209M parameters). The models give probabilistic, multivariate zero-shot forecasts and support past-only, known-future and categorical covariates. They run on CPU: 10 ms (5M) to 140 ms (209M)…

1d ago/1 source/medium confidence

CapabilityInfrastructureAdoption