GPT-6.1 Terra
OpenAI's mid-tier computer-use model lands at $1 per million input tokens — half the price of GPT-5.6 Terra and more than three times faster. — The price cut, not the benchmark, is the story.
AI Lately · 27 Tue · +1
№ 1 · Oct 24–30
One link a week. Everything that mattered in AI.
№ 1
OpenAI, then Anthropic hours later, say their models produced sub-exponential algorithms for a hard class of logic problems. Outside reviewers are now checking the proofs.
Anthropic filed to go public, disclosing $131.7 billion in cash and investments, and chip stocks rallied for a third consecutive session on the news.
OpenAI released GPT-6.1 Terra, a mid-tier computer-use model priced at half its predecessor, and said its models produced three mathematical results including a claimed refutation of the Exponential Time Hypothesis.
Within hours, Anthropic claimed a refutation of its own, leaving outside shepherds to decide whether either proof survives review.
Google DeepMind previewed Genie 4, a world model that generates interactive scenes with sound, while Meta open-sourced the 816-billion-parameter Muse Spark Open for agentic coding.
The money kept moving: Tenstorrent raised a $1.5 billion Series E at a $6.2 billion valuation, and Guidewire agreed to acquire hyperexponential for $1.7 billion in cash.
Underneath the rally, investors kept debating whether debt-funded AI capital expenditure can pay for itself.
OpenAI's mid-tier computer-use model lands at $1 per million input tokens — half the price of GPT-5.6 Terra and more than three times faster. — The price cut, not the benchmark, is the story.
AI Lately · 27 Tue · +1
Google DeepMind's world model generates interactive scenes with native audio and real-time steering, in research preview for trusted testers.
The Morning Build · 28 Wed
Meta's 816-billion-parameter open-weight MoE targets long-running coding agents, with 42 billion parameters active per token and a one-million-token context.
Creati AI · 27 Tue
Microsoft AI's latest image editing model debuts at No. 3 on Arena Edit and No. 2 on Artificial Analysis's image editing leaderboard.
Microsoft AI · 27 Tue
Six compact models — language, vision, speech, structured decisions — built to run enterprise agents on company-controlled hardware.
VentureBeat · 27 Tue
Company-reported score more than doubles its predecessor; no independent evaluation had been published by Monday night.
AI Atlas · 26 Mon
50.3%
Microsoft's image editor debuts third on Arena Edit, making Microsoft AI the No. 2 lab on the board.
Microsoft AI · 27 Tue
#3
Liquid's smallest text model claims browser-agent results above substantially larger rivals on selected tool-use tasks.
VentureBeat · 27 Tue
38.42
A proactive Anthropic agent that understands what needs your attention — 322 votes on Product Hunt this week.
Launch Index · 27 Tue
Lets AI agents sign in for you without ever seeing your passwords.
Launch Index · 24 Sat
Mistral's speech models for transcription, text-to-speech, and agents ship on Product Hunt.
Launch Index · 22 Thu
Turns live speech into translated audio as people talk.
Launch Index · 22 Thu
The filing discloses $131.7 billion in cash and investments; the paperwork alone was enough to lift AI stocks for a third straight session.
SiliconReport · 27 Tue · +1
$131.7B
Jim Keller's RISC-V AI compute company reaches a $6.2 billion valuation as it scales Galaxy server production and IP licensing.
Business Wire · 27 Tue
$1.5B
Specialized cloud infrastructure and chips for AI inference, at an $800 million valuation.
SiliconReport · 27 Tue
$100M
Autonomous AI agent systems for enterprise financial operations.
Funding Index · 27 Tue
$45M
Secrets-management infrastructure for enterprise systems and AI agents, at a $600 million valuation.
Funding Index · 27 Tue
$40M
The P&C insurer platform buys the London-based agentic underwriting specialist, setting up an 'agentic underwriting' push across commercial lines.
PR Newswire · 27 Tue
$1.7B
The managed AI-execution company absorbs the Autonomous Aftermarket pioneer and its Insyghts platform, a week after closing a $29M Series B.
GlobeNewswire · 28 Wed
undisclosed
OpenAI says its models found sub-exponential algorithms for a hard class of logic puzzles; Anthropic reportedly claimed its own refutation hours later. Outside shepherds are organising expert groups to verify each result.
AI Atlas · 26 Mon · +1
Relief rally or reckoning: Anthropic's IPO paperwork lifted chip stocks all week, while investors keep asking whether borrowing-fuelled data-centre spend earns its cost of capital.
Reuters · 27 Tue · +1
Data: Exa Markets · CBC Globe
| Ticker | Close | Δ week |
|---|---|---|
| NVDA | 217.94 | +3.31% |
| MSFT | 531.33 | +1.3% |
| GOOGL | 320.58 | +0.12% |
| META | 620.19 | -3.49% |
| AMD | 560.26 | +8.41% |
| TSM | 436.87 | +4.39% |
| AMZN | 253.1 | +2.06% |
| ARM | 226.6 | +7.73% |
| # | Company | Stories | Trend |
|---|---|---|---|
| 1 | OpenAI | 3 | |
| 2 | Anthropic | 3 | |
| 3 | Liquid AI | 2 | |
| 4 | Meta | 2 | |
| 5 | 1 |
The Weekly AI Edition
Cover
OpenAI, then Anthropic hours later, say their models produced sub-exponential algorithms for a hard class of logic problems. Outside reviewers are now checking the proofs.