Est. MMXXVI No. 78 ◆ Archive ◉ RSS ⌁ Free Models
LVX IN TENEBRIS

Inference server generously provided by Devocracy · powered by DwarfStar · DeepSeek V4 Flash 0731 (M3 Ultra)!

Corsair Discounts: Up to 50% Off Across Gaming Gear for September 2026 — wired.com
OpenAI Pushes Congress for Clarity on the Legality of an AI Industry Slowdown — wired.com
Collaborative Fund Bets on D.C. United as Investor Ownership of Sports Grows — techcrunch.com
Here is the article based only on the source text. — techcrunch.com
Slack is rolling out a feature that lets people build interactive tools directly inside their conversations, according to The Verge. Called Slackforce Surfaces, the capability lets users describe what they need to Slackbot, which uses AI to gather information from relevant conversations and connected apps, such as Google Drive or Salesforce, to create it. — theverge.com
Corsair Discounts: Up to 50% Off Across Gaming Gear for September 2026 — wired.com
OpenAI Pushes Congress for Clarity on the Legality of an AI Industry Slowdown — wired.com
Collaborative Fund Bets on D.C. United as Investor Ownership of Sports Grows — techcrunch.com
Here is the article based only on the source text. — techcrunch.com
Slack is rolling out a feature that lets people build interactive tools directly inside their conversations, according to The Verge. Called Slackforce Surfaces, the capability lets users describe what they need to Slackbot, which uses AI to gather information from relevant conversations and connected apps, such as Google Drive or Salesforce, to create it. — theverge.com

OPEN WEIGHTS

DeepSeek releases V4.1-Flash, a natively multimodal open-weights model

DeepSeek shipped V4.1-Flash with open weights on September 10 — a natively multimodal (vision and language) model built on a 552B sparse backbone with roughly 8B–16B active parameters. It supports up to 1M context, a shared compressed KV cache, a two-stage sparse indexer, and a 196B Engram lookup memory, with SGLang adding day-0 inference and reinforcement-learning support.

Research & Papers

05

CausalArena: benchmarking causal discovery in the foundation model era

Open Source & Models

03

AuK, an open-source foundation model for speech generation and editing

Hardware & Robotics

04

Skild AI taps NVIDIA physical AI stack to teach robots via S1 model

Tools & Startups

05

Desert Ant Labs: local, fast AI models that run entirely on device

Money & Markets

05

Arlequin AI lands EUR 28M Series A to scale topological neural network technology

Free Models

25

Free means $0 per token today, not unlimited or unconditional: both platforms rate-limit free use and require a free API key, some OpenRouter free endpoints may train on your prompts unless you opt out in account settings, and the roster changes daily. Verified against each provider's own pricing data at generation time — but pricing changes without notice, so confirm it's still free on the provider's own page before you rely on it.

In Brief

12

Source: